A Compendium of

Musical
Mathematics
This page intentionally left blank
 A Compendium of
Musical
Mathematics

Franck Jedrzejewski
CEA, Paris-Saclay University, France

World Scientific
NEW JERSEY • LONDON • SINGAPORE • BEIJING • SHANGHAI • HONG KONG • TAIPEI • CHENNAI • TOKYO
Published by
World Scientific Publishing Co. Pte. Ltd.
5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE

Library of Congress Cataloging-in-Publication Data

Names: Jedrzejewski, Franck, author.
Title: A compendium of musical mathematics / Franck Jedrzejewski,
Paris-Saclay University, France.
Description: New Jersey : World Scientific, 2024. | Includes bibliographical references and index.
Identifiers: LCCN 2023040815 | ISBN 9789811284366 (hardcover) |
ISBN 9789811284373 (ebook) | ISBN 9789811284380 (ebook other)
Subjects: LCSH: Music--Mathematics. | Musical intervals and scales.
Classification: LCC ML3800 .J43 2024 | DDC 780/.051--dc23/eng/20231205
LC record available at https://lccn.loc.gov/2023040815

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.

Cover: Parallel classes of block design (12, 4, 3) by Tom Johnson

Copyright © 2024 by World Scientific Publishing Co. Pte. Ltd.

All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage and retrieval
system now known or to be invented, without written permission from the publisher.

For any available supplementary material, please visit

https://www.worldscientific.com/worldscibooks/10.1142/13626#t=suppl

Desk Editors: Nimal Koliyat/Rok Ting Tan

Typeset by Stallion Press

Email: enquiries@stallionpress.com

Printed in Singapore
 About the Author

Franck Jedrzejewski is a researcher in math-

ematical physics at INSTN/CEA (Paris-Saclay
University). He has also a PhD in music and
musicology and in philosophy. In 2013, he
was elected Director of Program at College
International de Philosophie (CIPh), an institute
founded in 1983 by Jacques Derrida, François
Châtelet and Dominique Lecourt. He has pub-
lished more than 20 books, most of them in
French: Dictionnaire des musiques microtonales
(Dictionary of Microtonal Music), Hétérotopies musicales (Modèles
mathématiques de la musique), La musique dodécaphonique et
sérielle: une nouvelle histoire, or in English Looking at Numbers with
the composer Tom Johnson. His research is highly transdisciplinary
(music, philosophy, mathematics and Polish and Russian avant-
gardes). He currently teaches at Paris-Saclay University and the
National Institute for Nuclear Science and Technology (INSTN).

v
This page intentionally left blank
 Contents

About the Author v

Introduction xi

1. Musical Set Theories 1

1.1 Pitch Classes . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Chords and Scales . . . . . . . . . . . . . . . . . . . 2
1.3 Sets of Limited Transposition . . . . . . . . . . . . . 4
1.4 Enumeration of Chords and Scales . . . . . . . . . . 15
1.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . 28
References . . . . . . . . . . . . . . . . . . . . . . . 29
2. Generalized Interval Systems 31
2.1 Generalized Interval System . . . . . . . . . . . . . . 31
2.2 Interval Function . . . . . . . . . . . . . . . . . . . . 35
2.3 Injection Number . . . . . . . . . . . . . . . . . . . 38
2.4 Babbitt’s Hexachord Theorem . . . . . . . . . . . . 43
2.5 Interval Sum . . . . . . . . . . . . . . . . . . . . . . 44
2.6 Indicator Function . . . . . . . . . . . . . . . . . . . 47
2.7 Homometric Sets . . . . . . . . . . . . . . . . . . . . 48
2.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . 54
References . . . . . . . . . . . . . . . . . . . . . . . 56
3. Generalized Diatonic Scales 57
3.1 Sets of Progressive Transposition . . . . . . . . . . . 58
3.2 Well-Formed Scales . . . . . . . . . . . . . . . . . . 60

vii
viii A Compendium of Musical Mathematics

3.3 Generalized Diatonic Scales . . . . . . . . . . . . . . 65

3.4 Generalized Major and Minor Scales . . . . . . . . . 69
3.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . 74
References . . . . . . . . . . . . . . . . . . . . . . . 75
4. Voice Leading and Neo-Riemannian
Transformations 77
4.1 Isographic Networks . . . . . . . . . . . . . . . . . . 77
4.2 Automorphisms of the T /I Group . . . . . . . . . . 78
4.3 Automorphisms of the T /M Group . . . . . . . . . 81
4.4 PLR Transformations . . . . . . . . . . . . . . . . . 84
4.5 JQZ Transformations . . . . . . . . . . . . . . . . . 86
4.6 Neo-Riemannian Groups . . . . . . . . . . . . . . . 92
4.7 Atonal Triads . . . . . . . . . . . . . . . . . . . . . . 96
4.8 Seventh Chords . . . . . . . . . . . . . . . . . . . . 97
4.9 Hierarchy of Rameau Groups . . . . . . . . . . . . . 104
4.10 Exercises . . . . . . . . . . . . . . . . . . . . . . . . 105
References . . . . . . . . . . . . . . . . . . . . . . . 106
5. Combinatorics on Musical Words 109
5.1 Musical Words . . . . . . . . . . . . . . . . . . . . . 109
5.2 Syntactic Monoids . . . . . . . . . . . . . . . . . . . 111
5.3 Formal Grammars . . . . . . . . . . . . . . . . . . . 113
5.4 Words and Rhythms . . . . . . . . . . . . . . . . . . 114
5.4.1 Lyndon words . . . . . . . . . . . . . . . . . 114
5.4.2 Euclidean rhythms . . . . . . . . . . . . . . 115
5.4.3 Maximally even rhythms . . . . . . . . . . . 116
5.4.4 Deep rhythms . . . . . . . . . . . . . . . . . 117
5.4.5 Rhythmic oddity . . . . . . . . . . . . . . . 119
5.5 Words and Scales . . . . . . . . . . . . . . . . . . . 121
5.5.1 Dyck words . . . . . . . . . . . . . . . . . . 121
5.5.2 Infinite words and complexity . . . . . . . . 122
5.5.3 Sturmian words . . . . . . . . . . . . . . . . 123
5.5.4 Christoffel words . . . . . . . . . . . . . . . 125
5.5.5 Sturmian morphisms . . . . . . . . . . . . . 126
5.6 Plactic Congruences . . . . . . . . . . . . . . . . . . 132
5.6.1 Robinson–Schensted–Knuth
correspondence . . . . . . . . . . . . . . . . 133
5.6.2 Plactic modal classes . . . . . . . . . . . . . 135
 Contents ix

5.6.3 Pentatonic modes . . . . . . . . . . . . . . . 137

5.6.4 Hexatonic modes . . . . . . . . . . . . . . . 139
5.6.5 Heptatonic modes . . . . . . . . . . . . . . . 139
5.6.6 Octatonic modes . . . . . . . . . . . . . . . 140
5.7 Rational Associahedra . . . . . . . . . . . . . . . . . 141
5.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . 146
References . . . . . . . . . . . . . . . . . . . . . . . 147
6. Rhythmic Canons 151
6.1 Tilings . . . . . . . . . . . . . . . . . . . . . . . . . 152
6.1.1 Mask polynomials . . . . . . . . . . . . . . . 153
6.1.2 Cyclotomic polynomials . . . . . . . . . . . 154
6.1.3 Basic properties of tilings . . . . . . . . . . . 157
6.1.4 Perfect rhythmic tilings . . . . . . . . . . . . 160
6.2 Tijdeman’s Theorem . . . . . . . . . . . . . . . . . . 161
6.3 Hajós Groups . . . . . . . . . . . . . . . . . . . . . . 162
6.4 Coven–Meyerowitz Conjecture . . . . . . . . . . . . 164
6.5 Fuglede Conjecture . . . . . . . . . . . . . . . . . . 165
6.6 Vuza Canons . . . . . . . . . . . . . . . . . . . . . . 167
6.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . 174
References . . . . . . . . . . . . . . . . . . . . . . . 175
7. Serial Knots 177
7.1 Chord Diagrams . . . . . . . . . . . . . . . . . . . . 177
7.2 Enumeration of Tone Rows . . . . . . . . . . . . . . 179
7.3 All-Interval 12-Tone Rows . . . . . . . . . . . . . . . 182
7.4 Types of Tone Rows . . . . . . . . . . . . . . . . . . 183
7.5 Combinatoriality . . . . . . . . . . . . . . . . . . . . 185
7.6 Similarity Measures . . . . . . . . . . . . . . . . . . 187
7.7 Serial Groups . . . . . . . . . . . . . . . . . . . . . . 190
7.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . 193
References . . . . . . . . . . . . . . . . . . . . . . . 195
8. Combinatorial Designs 197
8.1 Difference Sets . . . . . . . . . . . . . . . . . . . . . 197
8.2 Block Designs . . . . . . . . . . . . . . . . . . . . . 199
8.3 Resolvable Designs . . . . . . . . . . . . . . . . . . . 204
8.4 Kirkman’s Ladies . . . . . . . . . . . . . . . . . . . 206
8.5 Block Design Drawings . . . . . . . . . . . . . . . . 206
x A Compendium of Musical Mathematics

8.6 Tom Johnson’s Graphs . . . . . . . . . . . . . . . . 209

8.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . 211
References . . . . . . . . . . . . . . . . . . . . . . . 211
9. Tuning Systems 213
9.1 Cents and Beats . . . . . . . . . . . . . . . . . . . . 213
9.2 Some Commas . . . . . . . . . . . . . . . . . . . . . 215
9.3 Historical Temperaments . . . . . . . . . . . . . . . 216
9.4 Harmonic Metrics . . . . . . . . . . . . . . . . . . . 218
9.5 Continued Fractions . . . . . . . . . . . . . . . . . . 220
9.6 Best Approximations . . . . . . . . . . . . . . . . . 223
9.7 Musical Scale Construction . . . . . . . . . . . . . . 228
9.7.1 Euler–Fokker genera . . . . . . . . . . . . . 230
9.7.2 Farey tunings . . . . . . . . . . . . . . . . . 231
9.7.3 Partch odd-limit . . . . . . . . . . . . . . . . 233
9.8 Three-Gap Theorem and Cyclic Tunings . . . . . . . 234
9.8.1 Cyclic tunings . . . . . . . . . . . . . . . . . 235
9.8.2 Ervin Wilson’s CPS . . . . . . . . . . . . . . 236
9.9 Tuning Theory . . . . . . . . . . . . . . . . . . . . . 237
9.9.1 Hellegouarch’s commas . . . . . . . . . . . . 237
9.9.2 Tuning systems . . . . . . . . . . . . . . . . 239
9.10 Exercises . . . . . . . . . . . . . . . . . . . . . . . . 242
References . . . . . . . . . . . . . . . . . . . . . . . 243

Solutions to Exercises 245

Index 265
 Introduction

The purpose of this book is to provide a concise introduction to

the mathematical theory of music, opening each chapter to the
most recent research. Despite the complexity of some sections, this
book can be read by a large audience. Many examples and exercises
illustrate the concepts introduced. This book is divided into nine
chapters. Each one stands as a separate essay.
In the first chapter, we tackle the question of the classification
of chords and scales. Some authors consider that two transposed
chords are in the same class, while others consider that they must
be linked by a transposition or an inversion to belong to the same
class. Still others evoke other relationships and other symmetries.
From a mathematical point of view, this way of looking at the
classification of chords and scales is the result of the (mathematical)
group action on the combinations of sounds. History has essen-
tially retained four groups: the cyclic group proposed by Anatol
Loquin (1874) and Edmond Costère (1954) which leads to a classifi-
cation up to transposition, the dihedral group proposed by Howard
Hanson (1960) and Allen Forte (1972) which leads to a classification
up to transposition and inversion, the affine group proposed by Czech
composer Aloı̈s Piňos (1971) and the symmetrical group proposed
by Mexican composer Julio Estrada (1994). But other groups are
possible. For the cyclic and dihedral groups, we show that Messiaen’s
limited transposition modes play an essential role in the classification
of chords and scales.
Chapter 2 is a mathematical presentation of David Lewin’s
Generalized Interval Systems and it is an application of the group

xi
xii A Compendium of Musical Mathematics

action seen in the previous chapter. This musicologist proposed to

generalize the notion of interval by linking it to a group and to a
space on which the group acts. The interval function makes it pos-
sible to establish very general results independently of the context.
The chapter ends with a study of homometric sets or Z relation,
proposed by Allen Forte. Two sets are in Z relation if they have the
same interval content. In other words, if they sound similar. Just as
Messiaen’s modes with limited transposition play an important role
in the classification of chords and scales, homometric sets form the
core of a classification by interval content.
Chapter 3 proposes a new theory of diatonicity in equal-
temperament universes. When we consider a space that is equally
divided into more than 12 notes, the question arises as to what a
diatonic scale is in this space (the white keys of the piano) or, in
other words, what would be the equivalents of the black and white
keys of the piano in N-tone equal temperament. Is there a diatonic
scale, and if so, can we construct an equivalent of the major and
minor scales? We offer a solution to these questions based on the
notions of well-formed scales and on the modes of progressive trans-
position introduced by French composer Alain Louvier. These modes
play a role similar to that of Messiaen’s limited transposition modes,
discussed in the first chapter.
Chapter 4 deals with the theory of voice leading and neo-
Riemannian transformations, based on the work of David Lewin,
Richard Cohn and Henry Klumpenhouwer. Can we explain har-
mony by an isographic network of chords? This is Klumpenhouwer’s
hypothesis. By introducing the Riemannian transformations P, L, R
whose name refers to the early work of German musicologist Hugo
Riemann, David Lewin attempted to map the evolution of major and
minor triads. However, these transformations depend on the context
and are only defined for the set of major and minor triads. This is
why we propose to introduce new noncontextual transformations J,
Q, Z which make it possible to follow the evolution of all types of
chords, from the atonal triad to seventh, ninth or other collections.
Chapter 5 is devoted to the application of word combinatorics to
music, which is another way of approaching musical structures such
as chords, scales and rhythms by reducing them to abstract words.
Dyck’s and Christoffel’s words represent musical scales and have
singular properties highlighted by Norman Carey, David Clampitt
 Introduction xiii

and Thomas Noll. This is another approach to diatonicity based

on Christoffel’s standard word factorization. Plactic congruence is
another way of classifying chords and scales. Introduced by Donald
Knuth, they find a natural equivalent in music. They lead to the
construction of the plactic monoid which allows a classification of
modes in all equal tempered universes. Rational associahedra give a
presentation associated with Dyck’s words and the generalization of
Catalan numbers.
Chapter 6 studies the rhythmic canons and the tessellation of
the line. This question is related to the problem of factoring finite
Abelian groups. It has been studied by many mathematicians like
N.G. de Bruijn, A.D. Sands and S. Szabo. Some canons such as
perfect canons or Vuza canons have remarkable properties, linked
with non-Hajós groups.
Chapter 7 is devoted to serial knots. The 12-tone rows invented
by J.M. Hauer and A. Schönberg can be classified using a concept of
knot theory called a chord diagram. Thanks to these structures, it is
possible to classify the 9 million 12-tone rows according to 554 chord
diagrams. We also present certain concepts such as the combinato-
riality introduced by the composer Milton Babbitt, some measures
between series studied by Tuukka Ilomäki and lastly, the character-
ization of the generation power of a given 12-tone rows by its serial
group.
Chapter 8 presents combinatorial designs and their applications to
music. These mathematical structures have the property that some
sets of numbers, assimilated to chords or rhythms, have the partic-
ularity of forming a system in which any pair of numbers appears
only a given number of times in all of these sets. They were studied
by composer Tom Johnson who sought to represent them by graphs
and to highlight their intrinsic relationships.
Chapter 9 concerns the study of tuning systems. Contemporary
systems which extend old historical temperaments have remarkable
properties. They have been explored by composers involved in micro-
tonal music and just intonation. They are also the basis of the
techniques of composition of spectral music which one finds in the
music of some composers, such as Tristan Murail, Hugues Dufourt
and Gérard Grisey.
As we have seen, this book is not about analysis or synthesis
of sound, signal processing or computer music, but it deals with
xiv A Compendium of Musical Mathematics

abstract realm musical structures (chords, scales, rhythms, tone rows,

etc.). The mathematical study of these structures leads to a better
understanding of their involvement in the musical fabric and in com-
positions. Thus, for instance, Messiaen’s limited transposition modes
appear naturally, as structures highly involved in the enumeration of
chords and scales. Most of these structures are involved in cutting-
edge research and give new interpretations of the complicated struc-
tures of contemporary or experimental music, as well as microtonal
music.
Finally, one may be surprised not to find any musical example
written in a traditional way on a staff, but the musico-mathematical
theories that we present are intended to be independent of all nota-
tions and are based on their own numerical notation. This does not
prevent us from referring to many musical scores.
To conclude, I would like to thank the composers who accompa-
nied me and always supported my research: Tom Johnson, Alain
Louvier and Hugues Dufourt. They have always been sources of
inspiration.
My thanks also go to my colleagues Emmanuel Amiot, Moreno
Andreatta, José Luis Besada, Gilbert Delor, Thomas Fiore, Harald
Fripertinger, Fabien Lévy, Thomas Noll, Athanase Papadopoulos,
Ludovico Pernazza and Alexandre Popoff, as well as PhD students.
My thoughts also go out to two theorists who recently passed
away, Yves Hellegouarch and Jack Douthett, for their work, their
encouragement and their friendship.
Finally, I thank mathematician Paul Sampson, who corrected
several errors, the anonymous reviewers for their wise advice and
the editor who did a remarkable job.
To go further, we invite the reader to consult the following works.

References

Benson, D. (2006). Music: A Mathematical Offering, Cambridge University Press,

Cambridge.
Douthett, J., Hyde, M.M. and Smith, C.J. (2008). Music Theory and Mathe-
matics, Chords, Collections and Transformations, University of Rochester
Press, Rochester, New York.
Forte, A. (1977). The Structure of Atonal Music, 2nd edn., Yale University Press,
New Haven.
 Introduction xv

Jedrzejewski, F. (2006). Mathematical Theory of Music, Ircam/Delatour, Paris.

Jedrzejewski, F. (2019). Hétérotopies musicales. Modèles mathématiques de la
musique, Hermann, Paris.
Lewin, D. (1987). Generalized Musical Intervals and Transformations, Yale
University Press, New Haven.
Mazzola, G., Göller, S. and Müller, S. (2002). The Topos of Music. Geometric
Logic of Concepts, Theory, and Performance, Birkhäuser, Basel.
Mazzola, G. et al. (2018). The Topos of Music, 4 Vols., Springer, Berlin.
This page intentionally left blank
 Chapter 1

Musical Set Theories

Classification of scales began to take shape in the 19th century

through the works of Camille Durutte, Hoëne Wronski, Anatol
Loquin (see Loquin, 1874, 1884) and some others, but it really took
a new start in the 20th century (see Clough et al., 1999). J.M. Hauer
examines the 66 hexatonic scales classification called tropes, recovery
by Barbour in 1949. Busoni (1907) produced one of the first taxon-
omy of chords and scales of seven notes, followed by A. Hába (1927)
and completely achieved by Costère in 1953 (see Costère, 1954, 1962).
Slonimsky wrote Thesaurus of Scales and Melodic Patterns in 1947.
A. Forte, A. Vieru and F. Lerdhal gave new insights to the classifica-
tion of scales. And at the beginning of the 20th century, most of the
jazz theorists were looking for a new paradigm. George Russel gave
a new Lydian Chromatic Concept of Tonal Organization in 1959 and
Y. Lateef wrote Repository of Scales and Melodic Patterns (1981).

1.1 Pitch Classes

Definition 1.1. A pitch is a single sound at a distinguishable

frequency.
In standard Western music, pitches are equivalent modulo the
octave. If x and y are pitches of different frequencies, x ∼ y if
x/y = 2n for some n is an equivalence relation (n is a whole number
of octaves). The equivalence classes are called pitch classes.

1
2 A Compendium of Musical Mathematics

Definition 1.2. Pitch classes are groups of pitches related by octave

and enharmonic equivalence.
Middle C is (usually) 440 Hz. Middle C is assigned the number 0,
the next pitch is 1 and so on. The pitch just below Middle C is −1,
the next one in decreasing order is −2, etc. In standard Western
music, there are 12 pitch classes C = 0, C = D = 1, D = 2,
D = E = 3, . . . , B = 11. Pitch classes are conveniently represented
on a circle.
Definition 1.3. The interval class (abbreviated ic) is the interval
between two pitch classes, as the shorter number of steps around the
circle from one pitch class to the other.
Each interval class has two tonal counterparts: ic 0 = unison (P1)
or octave (P8), ic 1 = minor 2nd (m2) and major 7th (M7), ic 2 =
major 2nd (M2) and minor 7th (m7), ic 3 = minor 3rd (m3) and
major 6th (M6), ic 4 = major 3rd (M3) and minor 6th (m6), ic 5 =
perfect 4th (P4) and perfect 5th (P5) and ic 6 = augmented 4th (A4)
and diminished 5th (d5).
Definition 1.4. A pitch-class set (pcset) is a set of pitch classes.
Definition 1.5. The frequency ratio of a pitch of frequency v rel-
atively to above C of frequency v0 is v/v0 . Frequency ratios are
expressed in cents

1200 log 2 (v/v0 )

In equal tempered tuning of N degrees by octave, the frequency

ratio of pitch class is 2n/N , for n = 0, 1, . . . , N − 1. In this case, the
arithmetic is modulo N , and the set of pitch classes is isomorphic to
ZN = Z/N Z.
Definition 1.6. A pc-set is a subset of the cyclic group ZN .

1.2 Chords and Scales

Definition 1.7. The function Tn : ZN → ZN , x → x + n mod N

is a bijection called transposition by musicians and translation by
mathematicians.
 Musical Set Theories 3

Example 1.1. T2 (2) = 4 means the transposition of D by a tone

is E.
Definition 1.8. The function In : ZN → ZN , x → −x + n mod N
is a bijection called inversion by musicians and reflection by math-
ematicians.
Example 1.2. I0 (4) = −4 = 8, the inversion of E is A, and I1 (4) =
−4 + 1 = 9, the inversion of E a semitone up is A.
The multiplication (considered by Robert Morris, 1991) Ma :
ZN → ZN , x → ax is not always a bijection. For N = 12, the only
values are a = 1, 5, 7, 11. Since M1 = T0 and M11 = I0 , there are only
two multiplications in the twelve-tone universe: M5 and M7 = I0 M5 .
All these transformations could be interpreted as permutations.
Examples are as follows:
 
0 1 2 3 4 5 6 7 8 9 10 11
I0 =
0 11 10 9 8 7 6 5 4 3 2 1
and
 
0 1 2 3 4 5 6 7 8 9 10 11
M5 =
0 5 10 3 8 1 6 11 4 9 2 7
Definition 1.9. A set class is a class, or general grouping, of pitch-
class sets that are understood to be functionally equivalent for some
purpose. A scale is a series of pitches in a distinct order (usually
ascending or descending) ordered by fundamental frequency or pitch.
Elements of a scale or elements of a mode are called degrees or notes.
Scales are named by their interval classes.
Usually, scales are defined by their interval sequence (e.g. major
scale 2212221, minor scale 2122122, etc.), gamuts are defined by their
transpositions (e.g. C major, D major, etc.) and modes are defined
by the cyclic permutations of interval sequence (e.g. Ionian 2212221,
Dorian 2122212, Phrygian 1222122, etc. are rotations of the major
scale). For musicologists, a mode is a type of musical scale coupled
with a set of characteristic melodic behaviors.
Example 1.3. In a chromatic scale, each scale step represents a
semitone interval. A major scale is defined by the interval pat-
tern 2212221. A (natural) minor scale has step intervals 2122122
4 A Compendium of Musical Mathematics

(e.g. A-minor is A, B, C, D, E, F, G, A). A harmonic minor scale

is 2122131 (e.g. A-harmonic minor is A, B, C, D, E, F, G, A). The
whole tone scale on C is 222222 W = {C, D, E, F, G, A}. It has
only two transpositions W and T1 (W ).
Example 1.4. An octatonic scale is a scale of cardinal 8 (8 notes).
A pentatonic scale is a scale of cardinal 5 (5 notes). Major pentatonic
(or Ionian mode) is defined by the interval pattern 22323 (C, D, E,
G, A, C). Minor pentatonic (or Aeolian mode) is 32232 (A, C, D, E,
G, A).
Example 1.5. The interval patterns of modern church modes are
cyclic permutation of the major scale. Ionian 2212221, Dorian
2122212, Phrygian 1222122, Lydian 2221221, Mixolydian 2212212,
Aeolian 2122122 and Locrian 1221222.
Definition 1.10. A chord is any set of pitches (frequencies) consist-
ing of multiple notes that are heard as if sounding simultaneously.
A k-chord is a chord of cardinal k.
From the combinatorial point of view, there is no difference
between chords and scales. In practice, a scale has usually more than
five notes. Triads are 3-chords.
Example 1.6. A major triad (major chord) is a chord that has a
root, a major third and a perfect fifth, e.g. C major is {C, E, G} =
{0, 4, 7}.
Example 1.7. A minor triad (minor chord) is a chord that has a
root, a minor third and a perfect fifth, e.g. C minor is {C, E, G} =
{0, 3, 7}.

1.3 Sets of Limited Transposition

Sets of limited transposition are of first importance in the combina-

torial study of the N -tone equal temperament (N -TET) also bet-
ter known as N -equal division of the octave (N -EDO). In general,
each chord has N different transposition sets, but in some cases, the
number of transpositions is less than N . For instance, the tritone
 Musical Set Theories 5

(diabolus in musica) has only 6 transpositions, while other intervals

have 12 transpositions.
Definition 1.11. A pcset A is a set of limited transposition if

∃ n ∈ ZN \{0}, Tn (A) = A

In other words, if the isotropy group of ZN for the translations

HA = {n ∈ ZN , Tn (A) = A}

is not reduce to {0}. Let L be the set of the sets of limited transpo-
sition.
For enumeration purpose, we consider class of sets of limited
transposition.
Definition 1.12. Let L be the quotient set L = L/∼ of the set class
of limited transposition by the cyclic group, up to transpositions.
A generator set is a set of elements of the quotient set L contain-
ing {0}.
This choice is well defined because if A is a set of limited trans-
position (A ∈ L), then all its transpositions belong to L:

∀j Ti (Tj (A)) = Tj (Ti (A) = Tj (A)

The isotropy group HA is cyclic. So it is generated by an element i0 .

Proposition 1.1. Let A ∈ L such that Tn (A) = A for some n. If
a ∈ A, then

∀m ∈ ZN , Tmn (a) ∈ A

In other words, if a is a note of the limited transpositions pcset A,

then a + n, a + 2n, . . . , a + mn are also in A.
Proof. We have

Tmn (a) ∈ Tmn (A) = T(m−1)n (A) = · · · = A


6 A Compendium of Musical Mathematics

Theorem 1.1. The number of transpositions of the set A ∈ L is the

nonnegative integer k such that
k × |HA | = N
where N is the cardinal of the set ZN . In particular, the order of the
isotropy group HA is a divisor of N.
Proof. As k is by definition the cardinal of the orbit TA , k is also
the index of HA in the group of transformations G isomorphic to ZN .
This is in fact Lagrange’s theorem: the order of any finite group is
divisible by the order of any subgroup of it. 
Corollary 1.1. Let A be a set class of limited transposition contain-
ing {0} and HA its isotropy group, each generator set is the union of
A and sets of the form Tj (A) for j ∈ HA .
Proof. Let
 

k
B =A∪ Tji (A)
i=1
be a such set. Since A is invariant under Ti and
∀ji , Ti (Tji (A)) = Tji (Ti (A)) = Tji (A)
the set B is invariant under Ti . So B is in L. 
Theorem 1.2. If N is a prime number, the set of the sets of lim-
ited transposition is trivial, i.e. reduces to one element, namely the
chromatic scale.
Proof. It is well known that the subgroups of ZN are kZN , where
k is a divisor of N . If N is prime, the subgroups are reduced to {0}
and ZN itself. Thus, the isotropy group is trivial. 
Definition 1.13. Let x be a k-chord. The interval structure of x =
{x1 , . . . , xk } where x1 < · · · < xn is the vector [Δx1 , . . . , Δxk ] also
written as Δx1 · · · Δxk , where Δxi = xi+1 − xi for i = 1, . . . , k − 1
and Δxk = x1 − xk mod N .
Theorem 1.3. A pcset x is a limited transposition set if and only if
the interval structure [x] is a partition of N which has a symmetry of
order d, i.e. such that [x] = [A, . . . , A] where the subset A is repeated
d times and |x| = d. |A|.
 Musical Set Theories 7

Example 1.8. The set x = {0, 1, 3, 6, 7, 9} has a structure [x] =

[1, 2, 3, 1, 2, 3] = 123123. This structure has a 2-symmetry [x] =
[A, A] with A = [1, 2, 3].
In the first half of the 20th century, Olivier Messiaen introduced
the concept of modes of limited transposition (see Messiaen, 1944,
1996), but he only used some of them in his own compositions. Some
sets were already known before Messiaen presented his theory. The
whole tone scale was used by many composers. Claude Debussy wrote
a prelude (Voiles) using only this mode. Messiaen’s third mode is the
enneaphonic scale that Scriabin used in his Piano Prelude, opus 74.
An enumeration shows that there are 16 sets of limited transposition
whose classes (up to transpositions) are presented in Table 1.1.
Table 1.1 results from the following theorem established by
Bernard Parzysz in 1983 (Parzysz, 1984).
Theorem 1.4 (Parzysz, 1983). In the 12-EDO, up to transposi-
tions, there are only 16 sets of limited transposition which can be
classified by their numbers of transpositions k. (1) If k = 1, there is
only one set of limited transposition, (2) if k = 2, there is only one
set of limited transposition, (3) if k = 3, there are two sets of limited

Table 1.1. Sets of limited transposition.

Names Limited Transposition Sets Tr.

Tritone (0, 6) 6
Aug. Fifth (0, 4, 8) 4
Dim. Seventh (0, 3, 6, 9) 3
Alt. Seventh (0, 2, 6, 8) 6
Double Fifths (0, 1, 6, 7) 6
Mode 1 (0, 2, 4, 6, 8, 10) 2
Mode 2 (0, 1, 3, 4, 6, 7, 9, 10) 3
Mode 3 (0, 1, 2, 4, 5, 6, 8, 9, 10) 4
Mode 4 (0, 1, 2, 3, 6, 7, 8, 9) 6
Mode 5 (0, 1, 2, 6, 7, 8) 6
Mode 6 (0, 1, 2, 4, 6, 7, 8, 10) 6
Mode 7 (0, 1, 2, 3, 4, 6, 7, 8, 9, 10) 6
Mode A (0, 1, 4, 5, 8, 9) 4
Mode B (0, 1, 3, 6, 7, 9) 6
Mode C (0, 1, 4, 6, 7, 10) 6
Chromatic Scale (0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11) 1
8 A Compendium of Musical Mathematics

transposition, (4) if k = 4, there are three sets of limited transposi-

tions and (5) if k = 6, there are nine sets of limited transposition.
Proof. The 12-EDO is identified with Z12 . Since the order of the
isotropy group is a divisor of 12, we have to study five cases.
(1) If |HA | = 12, then the chromatic scale
E1 = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11}
is the only set of limited transposition.
(2) If |HA | = 6, then the isotropy group is the whole tone scale
HA = E2 = {0, 2, 4, 6, 8, 10}. With T1 (E2 ), we can make by union
the set E2 ∪ T1 (E2 ) = E1 which is already obtained.
(3) If |HA | = 4, then the isotropy group is the diminished seventh
chord HA = E3 = {0, 3, 6, 9}. By making the unions of E3
with T1 (E3 ) and T2 (E3 ), three new sets are obtained. The set
E3 ∪ T1 (E3 ) = E4 = {0, 1, 3, 4, 6, 7, 9, 10} is Messiaen’s mode 2.
The set E3 ∪T2 (E3 ) = {0, 2, 3, 5, 6, 8, 9, 11} = T2 (E4 ) is the trans-
position of E4 by 2. The set E3 ∪T1 (E3 )∪T2 (E3 ) is the chromatic
scale already obtained.
(4) If |HA | = 3, then the isotropy group is the augmented fifth
triad HA = E5 = {0, 4, 8}. Three sets are formed by unions
with Tj (E5 ). The set E5 ∪ T1 (E5 ) = E6 = {0, 1, 4, 5, 8, 9} is the
mode A. But the set E5 ∪T3 (E5 ) = T3 (E6 ) is the transposition of
the previous set. The union E5 ∪ T2 (E5 ) = E2 is the whole tone
scale. The set E5 ∪T1 (E5 )∪T2 (E5 ) = E7 = {0, 1, 2, 4, 5, 6, 8, 9, 10}
is Messiaen’s mode 3. The other combinations are two transpo-
sitions of this mode, namely E5 ∪ T1 (E5 ) ∪ T3 (E5 ) = T3 (E7 ) and
E5 ∪ T2 (E5 ) ∪ T3 (E5 ) = T2 (E7 ).
(5) If |HA | = 2, then the isotropy group is the tritone HA = E8 =
{0, 6}. The set
E8 ∪ T1 (E8 ) ∪ T2 (E8 ) ∪ T3 (E8 ) ∪ T4 (E8 )
= E9 = {0, 1, 2, 3, 4, 6, 7, 8, 9, 10}
is Messiaen’s mode 7. The other combinations are transpositions of
this mode
E8 ∪ T1 (E8 ) ∪ T2 (E8 ) ∪ T3 (E8 ) ∪ T5 (E8 ) = T5 (E9 )
E8 ∪ T1 (E8 ) ∪ T2 (E8 ) ∪ T4 (E8 ) ∪ T5 (E8 ) = T4 (E9 )
 Musical Set Theories 9

E8 ∪ T1 (E8 ) ∪ T3 (E8 ) ∪ T4 (E8 ) ∪ T5 (E8 ) = T3 (E9 )

E8 ∪ T2 (E8 ) ∪ T3 (E8 ) ∪ T4 (E8 ) ∪ T5 (E8 ) = T2 (E9 )

The set

E8 ∪ T1 (E8 ) ∪ T2 (E8 ) ∪ T3 (E8 ) = E10 = {0, 1, 2, 3, 6, 7, 8, 9}

is Messiaen’s mode 4. The following unions are transposition sets:

E8 ∪ T1 (E8 ) ∪ T2 (E8 ) ∪ T5 (E8 ) = T5 (E10 )

E8 ∪ T1 (E8 ) ∪ T4 (E8 ) ∪ T5 (E8 ) = T4 (E10 )
E8 ∪ T3 (E8 ) ∪ T4 (E8 ) ∪ T5 (E8 ) = T3 (E10 )

The set

E8 ∪ T1 (E8 ) ∪ T2 (E8 ) ∪ T4 (E8 ) = E11 = {0, 1, 2, 4, 6, 7, 8, 10}

is a new limited transposition set (Messiaen’s mode 6). The other

sets are transpositions of mode 6:

E8 ∪ T1 (E8 ) ∪ T3 (E8 ) ∪ T5 (E8 ) = T5 (E11 )

E8 ∪ T2 (E8 ) ∪ T4 (E8 ) ∪ T5 (E8 ) = T4 (E11 )
E8 ∪ T2 (E8 ) ∪ T3 (E8 ) ∪ T4 (E8 ) = T2 (E11 )

The sets

E8 ∪ T1 (E8 ) ∪ T3 (E8 ) ∪ T4 (E8 ) = E4

E8 ∪ T2 (E8 ) ∪ T3 (E8 ) ∪ T5 (E8 ) = T2 (E4 )

are transpositions of the mode 2 already obtained. The study of the

combinations of the form E8 ∪ Ti (E8 ) ∪ Tj (E8 ) reveals three new
limited transposition sets (not listed by Messiaen). The set E8 ∪
T1 (E8 ) ∪ T2 (E8 ) = E12 = {0, 1, 2, 6, 7, 8} is the mode 5. The following
10 A Compendium of Musical Mathematics

sets are transposition sets of mode 5:

E8 ∪ T1 (E8 ) ∪ T5 (E8 ) = T5 (E12 )

E8 ∪ T4 (E8 ) ∪ T5 (E8 ) = T4 (E12 )

The set E8 ∪ T1 (E8 ) ∪ T3 (E8 ) = E13 = {0, 1, 3, 6, 7, 9} is the mode B.

The following sets are transposition sets of mode B:

E8 ∪ T2 (E8 ) ∪ T5 (E8 ) = T5 (E13 )

E8 ∪ T3 (E8 ) ∪ T4 (E8 ) = T3 (E13 )

The set E8 ∪T1 (E8 )∪T4 (E8 ) = E14 = {0, 1, 4, 6, 7, 10} is the mode C.
The following sets are transposition of the preceding set:

E8 ∪ T3 (E8 ) ∪ T5 (E8 ) = T5 (E14 )

E8 ∪ T2 (E8 ) ∪ T3 (E8 ) = T2 (E14 )

The set E8 ∪ T2 (E8 ) ∪ T4 (E8 ) = E2 is the whole tone scale already

obtained. The next step is the study of the sets E8 ∪ Tj (E8 ). This
leads to two new limited transposition sets E8 ∪ T1 (E8 ) = E8 ∪
T5 (E8 ) = E15 = {0, 1, 6, 7} and E8 ∪ T2 (E8 ) = E8 ∪ T4 (E8 ) = E16 =
{0, 2, 6, 8}. The set E8 ∪ T3 (E8 ) = E3 is already obtained. 
A presentation of the sets of limited transposition is to put the
number of notes of the set on the x-axis and the number of transpo-
sitions on the y-axis (see Fig. 1.1). The five first sets are classified or
altered chords, the seven following modes (mode 1 to mode 7) show
Messiaen’s modes and the last three (modes A, B and C) are defec-
tive as they are included in Messiaen’s modes: mode A (or mode 8) in
mode 3, mode B (or mode 9) in mode 2 and mode C (or mode 10) is
included in mode 6. The chromatic scale has only one transposition
and contains all the other modes.
Remark the proof of Parzysz is tedious as N increases. Parzysz’s
theorem could be generalized to the quarter-tone universe, but the
proof is quite long, see Jedrzejewski (2006). We get the following
result:
Theorem 1.5. In the quarter-tone universe (24-EDO), there are
only 381 limited transposition sets which can be classified by the num-
ber of transpositions k. (1) If k = 1, there is only one set of limited
 Musical Set Theories
Fig. 1.1. Sets of limited transposition in the 12-EDO.

11
12 A Compendium of Musical Mathematics

transposition, (2) if k = 2, there is only one set of limited transpo-

sition, (3) if k = 3, there are two sets of limited transposition, (4)
if k = 4, there are three sets of limited transposition, (5) if k = 6,
there are 9 sets limited transposition, (6) if k = 8, there are 30 sets of
limited transposition and (7) if k = 12, there are 335 sets of limited
transposition.
The number of limited transposition sets for a given equal temper-
ament of N notes can be established in another way. Pólya theorem
gives of xk is exactly the number k -chord in N -tone equal tempera-
ment. For the action of the cyclic group, the polynomial is given by
the following formula:
 
1 N
P (x) = ϕ (1 + xN/d )d
N d
d|N

where ϕ is the Euler’s function. For x = 1, this formula gives the

number of pcset classes
 
1 d N
PN = 2 ϕ
N d
d|N

The following result has been established by Fabien Ballon in 2004

(unpublished). We denote LN the number of limited transposition
sets, MN the number of modes and KN the number of modes asso-
ciated with limited transposition sets. A pcset class of k degrees is
associated with at most k modes.
Theorem 1.6 (Ballon, 2004). The number of limited transposi-
tion sets is given by
2
LN = PN + (KN − MN )
N
where PN is given by the Pólya theorem,
 
1 d N
PN = 2 ϕ
N d
d|N

MN is the number of modes MN = 2N −1 and KN is given by

1  r−k+2
 
r r−k+1 r N
KN = (−1)k+1 ··· 2 pi1 pi2 ···pik
2
k=1 i1 =1 i2 =i1 +1 ik =ik−1 +1
 Musical Set Theories 13

where the decomposition in prime factors of N is N = pk11 pk22 · · · pkr r ,

with pi > 1, ki > 0 (i = 1, . . . , r) and r > 0.

Proof. Let PN,k be the number of pcset classes of k notes and

MN,k the number of modes of k notes. We have


N 
N
PN = PN,k and MN = MN,k
k=1 k=1

Let LN,k be the number of limited transposition set of k notes and

KN,k the number of modes associated with the limited transposition
sets of k notes. A pcset class of k notes which is not a limited trans-
position set has exactly k modes, thus the numbers of the respective
quantities verify the relation

1
PN,k − LN,k = (MN,k − KN,k )
k
As the number of pcset classes of k notes and the number of pcset
classes of N − k notes are equal, the sum over k leads to

2
PN − LN = (MN − KN )
N
From this relation, we get the expression of LN . On the other hand,
the number of modes of k notes is
 
N −1 (N − 1)!
MN,k = =
k−1 (N − k)!(k − 1)!

This expression leads to the total number of modes:


N 
N −1 
N −1
MN = MN,k = = 2N −1
k
k=1 k=0

The relation KN is more difficult to prove. Let MN be the set of

modes and KN the set of modes associated with limited transposition
sets. Let N be N = pk11 pk22 · · · pkr r and m be a divisor of N such
that m = 1. The elements of KN are constructed from the elements
14 A Compendium of Musical Mathematics

of the sets MN/pj with j = 1, . . . , r. For each divisor m of N , the

injection

fm : MN/m → KN

maps a mode M of N/m notes with a limited transposition modes

of KN by repeating N/m times the structure of M . The set KN is
the union of the sets

r 
KN = fpj MN/pj
j=1

By applying the following result to a collection of r sets A1 , . . . , Ar

and by denoting |A| the cardinal of the set A,


r 
r  r−k+2
r−k+1  
r
Aj = k+1
(−1) ··· Ai1 ∩ Ai2 ∩ · · · ∩ Aik
j=1 k=1 i1 =1 i2 =i1 +1 ik =ik−1 +1

we get


r  r−k+2
r−k+1 
KN = |KN | = (−1)k+1 ···
k=1 i1 =1 i2 =i1 +1


r
fpi1 ··· pik (MN/pi1 ··· pi )
k
ik =ik−1 +1

For two divisors m1 and m2 of N , the function fm verifies

fm1 (MN/m1 ) ∩ fm2 (MN/m2 ) = fm1 m2 (MN/m1 m2 )

But fm is an injective mapping. For all divisor m of N , we have

fm (MN/m ) = MN/m = MN/m

and then
N
fpi1 ..pik (MN/pi1 ..pi ) = MN/pi1 ..pi = 2 p i1 p i2 · · · p ik
k k

Replacing this relation in the expression of KN leads to the

result. 
 Musical Set Theories 15

The same demonstration can be used for the enumeration of the

modes of k notes associated with transposition limited sets.

Theorem 1.7. Let N be the number of notes per octave in the equal
temperament and k an integer less than N . Let’s denote m the num-
ber of common prime factors of N and k, and q1 , q2 , . . . , qm these
factors. The number of modes MN,k is given by


m  m−k+2
m−k+1  
m  N
− 1
qi1 qi2 ···qik
MN,k = (−1)k+1
···
k=1 i1 =1 i2 =i1 +1 ik = ik−1 +1
k
qi1 qi2 ···qik −1

By convention, the sums with terms in reverse order are ignored.

The number of pcset classes (Pcsc) in the N -EDO and the number
of limited transposition set classes (Ltsc) are given in the following
table for different values of N (in the cyclic classification).
For divisions of m-th of tone (m > 5, m = N/6), we get large
values.

1/m N Ltsc Pcsc

1/5 30 2 300 35 792 567
1/6 36 14 939 1 908 881 899
1/8 48 703 331 5 864 062 367 251
1/12 72 1 909 580 799 65 588 423 374 144 427 519
1/16 96 5 864 196 582 931 825 293 359 523 589 782 053 586 451

1.4 Enumeration of Chords and Scales

One might think that it is easy to count the number of k-chords in a

temperament equal to N degrees and that this number is simply the
number of k -combinations in a set of N elements, i.e. the coefficients
of the binomial
 
N N!
=
k k!(N − k)!

But this is wrong. The presence of sets of limited transposition (SLT)

complicates the calculation: if N is not a prime number, the number
16 A Compendium of Musical Mathematics

of k-chords up to transpositions is not a multiple of N . Let us look

at some examples:
• N = 12, k = 2, there is only one SLT: the tritone {0, 6} which has
6 transpositions, the other 5 intervals have 12 transpositions:
 
12
= 66 = 5 × 12 + 1 × 6
2
So there are only 5 + 1 = 6 intervals, up to transpositions.
• N = 12, k = 3, there is only one SLT: the triad {0, 4, 8} which has
4 transpositions:
 
12
= 220 = 18 × 12 + 1 × 4
3
So there are 18 + 1 = 19 triads, up to transpositions.
• N = 12, k = 4, there are 3 SLT: {0, 3, 6, 9} with 3 transpo-
sitions, {0, 2, 6, 8} with 6 transpositions and {0, 1, 6, 7} with
6 transpositions:
 
12
= 495 = 40 × 12 + 1 × 3 + 1 × 6 + 1 × 6
4
So there are 40 + 1 + 1 + 1 = 43 quadrichords up to transpositions.
In fact, we will see that the enumeration is facilitated by the use
of action groups and combinatorial results due to Burnside and Pólya
and does not require knowledge of sets of limited transposition, either
they are of some importance for the enumeration.
The purpose of the following is to propose a general framework
of chords’ classification using groups’ action and to rediscover the
cyclic classification proposed by Edmond Costère (1954), the dihedral
classification proposed by Howard Hanson (1946), Zalewski and Allen
Forte (1972), the affine classification proposed by the Czech composer
Aloı̈s Piňos (1971), and later by Daniel Starr (1978) and Guerino
Mazzola (1985), and the symmetric classification proposed by the
Mexican composer Julio Estrada (1994).
There are (at least) two main usual ways to classify chords:
(1) Up to transpositions: This was proposed by Anatole Loquin
(1834–1903) (see Loquin, 1874, 1884) and then later by Edmond
Costère (pseudonym of the famous magistrate of the French
 Musical Set Theories 17

Supreme Court Edouard Coester (1905–2001), for the most

part unknown) in Lois et styles des harmonies musicales
(Costère, 1954). Two chords A and B of same cardinality are
equivalent if it exists n, such that B = Tn (A).
(2) Up to transpositions and inversions: This was proposed by
Howard Hanson (1896–1981) in Harmonic Materials of Modern
Music (Hanson, 1946) and taken up by Allen Forte (1926–2014)
in The Structure of Atonal Music (1977). Two chords A and B of
same cardinality are equivalent if it exists n such that B = Tn (A)
or B = In (A).
Most of the results have been established by Harald Fripertinger
(1992). For an introduction to enumeration, one can consult Aigner
(2007).
Definition 1.14. A group action of a multiplicative group G on a
set X is given by a mapping G×X on X, (g, x) → g ·x which satisfies
the following conditions:
(1) 1.x = x, for all x ∈ X and where 1 is the unit element of G,
(2) (gh) · x = g(h · x), for all g, h ∈ G and x ∈ X.
The group action induces an equivalence relation x ∼ y given by
∃ g ∈ G, y = gx
The orbits are the equivalence classes of this relation. Two elements
x, y of X are in the same orbit if there is an element g of the group
G such that y = gx.
Definition 1.15. The orbits of G on X are the equivalence classes
G(x) of the previous equivalence relation
G(x) = {gx | g ∈ G}
The set of all orbits is denoted by G\\X = {G(x) | x ∈ X}. The
action is transitive (or G acts transitively) if there is only one orbit.
Definition 1.16. For each x ∈ X, the stabilizer Gx of x is the set
Gx = {g ∈ G | gx = x}
The set of all fixed points of g ∈ G is denoted by
Xg = {x ∈ X | gx = x}
18 A Compendium of Musical Mathematics

Example 1.9. A subgroup H of G acts by multiplication from the

right on G. The orbit H(g) is the left coset gH. In other words, two
elements g, g ∈ G are in the same left coset of H if there is an
element h ∈ H such that g = hg. The left coset is denoted by G/H
or gH

G/H = {gh | h ∈ H}

The set of the right cosets is denoted by H\G or Hg.

Example 1.10. A group acts on itself by conjugation G × G → G,
(g, h) −→ ghg−1 . The orbit

G(h) = {ghg−1 | g ∈ G}

is the conjugacy class of h. The stabilizer of h is the set of its com-

mutative elements

Gh = {g ∈ G, gh = hg}

Lemma 1.1. Let H be a subgroup of G. There is a bijection between

the right and the left cosets. In particular, the sets have the same
cardinality

|G/H| = |H\G|

Proof. Let g1 and g2 be two elements of G. We have

Hg1 = Hg2 ⇐⇒ g1 (g2 )−1 ∈ H

⇐⇒ g1 (g2 )−1 ∈ H
⇐⇒ (g1−1 )−1 .(g2 )−1 ∈ H
⇐⇒ g1−1 H = g2−1 H

The map Hg −→ g −1 H is an embedding from G/H in H\G. It is

onto because all classes gH can be rewritten as (g −1 )−1 H. 

Definition 1.17. Let H be a subgroup of G. The index of H in

G denoted by |(G : H)| is the cardinal of the left cosets (and of the
right cosets)

|(G : H)| = |G/H| = |H\G|

 Musical Set Theories 19

Definition 1.18. The index of the trivial subgroup H = {1} is called

the order of G
|(G : 1)| = |G|
Definition 1.19. A subgroup H of G is normal if the right cosets
are equal to the left cosets
∀g ∈ G, gH = Hg
Theorem 1.8 (Lagrange). The order and the index of a subgroup
H of a finite group G are divisors of the order of group G
|(G : H)| = |G| / |H|
Proof. The map h −→ gh is a bijection from H to gH. All classes
gH form a partition of distinct subsets of the number of the left
cosets. But the number of left cosets is the index of H in G, so
|G| = |(G : H)| . |H|


Theorem 1.9. The order of the orbit of x is equal to the index of

the stabilizer of one of its elements
|G(x)| = |G : Gx |
Proof. The map sending the left coset gH to the element gx ∈ X
is a bijection between the left cosets of H in G and the orbit of x. 
Theorem 1.10. A group action of a finite group G on a set X
induces a group homomorphism from G to the symmetric group SX
by g → g where g is the mapping x → gx which is called a permuta-
tion representation of G on X.
Lemma 1.2 (Burnside). Let G be a multiplicative finite group act-
ing on a finite set X. The number of G-orbits is the average number
of fixed points
1 
|G\\X| = |Xg |
|G|
g∈G

where Xg is the set of all fixed points of g ∈ G

Xg = {x ∈ X, gx = x}
20 A Compendium of Musical Mathematics

Proof.
 1 1  |G| 1 
|G\\X| = = = |Gx |
|G(x)| |G| |G(x)| |G|
x∈X x∈X x∈X
1   1   1 
= 1= 1= |Xg |
|G| |G| |G|
x∈X g∈Gx g∈G x∈Xg g∈G

In order to introduce Pólya’s enumeration theorem, we need a
generalized version of this lemma. Let R be a commutative ring such
that Q is a subring of R and G be a finite group acting on a finite
set X. A weight function is a function w : X → R constant on each
G-orbit, i.e.

w(gx) = w(x) ∀g ∈ G, ∀x ∈ X

The weight of an orbit G(x) is defined as the weight of any element

of this orbit w(G(x)) = w(x). We recover the original version of this
lemma by setting w(x) = 1 for all x ∈ X.

Lemma 1.3 (Burnside). The sum of weights of G-orbits is the

average number of weighted fixed points
 1  
w(u) = w(x)
|G|
u∈G\\X g∈G x∈Xg

Proof.
   w(u)   w(u)
w(u) = =
|u| |G(x)|
u∈G\\X u∈G\\X x∈u u∈G\\X x∈u

1  |G| 1 
= w(x) = |Gx | w(x)
|G| |G(x)| |G|
x∈X x∈X
1   1  
= w(x) = w(x)
|G| |G|
x∈X g∈Gx g∈G x∈Xg

In the following, X, Y are two finite sets, R is a commutative
ring containing Q as a subring and G is a group acting on X.
The set of configurations Y X is the set of functions from X to Y.
 Musical Set Theories 21

The group action of G on X induces an action of G on the set Y X

by G × Y X → Y X ,

(g, f ) → f ◦ g −1

where g is the permutation representation of g acting on X and the

weight function h : Y → R induces a weight function on Y X by
ω :YX →R

ω(f ) = h(f (x))

x∈X

This function ω is constant on the G-orbits on Y X . For each g ∈ G,

ω(gf ) = h(f (g −1 x)) = h(f (x)) = ω(f )

x∈X x∈X

Definition 1.20. Let G be a group acting on a finite set X. The

cycle index of this action is a polynomial P (G, X) of Q[t1 , . . . , t|X| ]
defined by

1 
|X|
j (g)
P(G,X) (t1 , . . . , t|X| ) = tkk
|G|
g∈G k=1

where jk (g) is the number of cycles of length k of the permutation g

in its decomposition as a product of independent cycles.

Theorem 1.11 (Pólya). The sum of weights of G-orbits on Y X is

given by

⎛ ⎞jk (g)
 1 
|X|

ω(u) = ⎝ h(y)k ⎠
X
|G|
u∈G\\Y g∈G k=1 y∈Y

where jk (g) is the number of cycles of length k of the permutation g

in its decomposition as a product of independent cycles.
22 A Compendium of Musical Mathematics

Proof. According to the generalization of Burnside’s lemma, we

have to compute over the fixed point function f the sum
 
ω(f ) = h(f (x))
f ∈YgX f ∈YgX x∈X

But f is constant on the cycles of the permutation representation

of g
 
ω(f ) = h(f (u))|u|
f ∈YgX f ∈YgX u∈<g>\\X

= h(y)|u|
u∈<g>\\X y∈Y
⎛ ⎞jk (g)
|X|

= ⎝ h(y)k ⎠
k=1 y∈Y


Since conjugate elements have the same number of k -cycles in

their decomposition, it is sufficient to compute the cycle index over
the conjugacy classes

1 
|X|
j (g c )
P(G,X) (t1 , . . . , t|X| ) = |c| tkk
|G|
c∈C k=1

where C is the set of all conjugacy classes. If for the weights of the
elements of Y we take power of an independent variable z, the power
series in Pólya’s formula are called the configuration counting series
and the figures counting series. The coefficient of order k of the first
series is the number of classes in Y X of weight z k , and the coefficients
of the second series are the numbers of elements of Y of weight z k .

Example 1.11. Let G = S3 be the group of permutation of the set

{1, 2, 3}. S3 has six elements s1 =(123)=(1)(2)(3), s2 =(132)=(1)(23),
s3 =(213)=(12)(3), s4 =(321)=(13)(2), s5 =(231)=(123) and s6 =
(312)=(123). For each element, the number of cycle of length k is
j1 (s1 ) = 3, j2 (s1 ) = 0, j3 (s1 ) = 0, j1 (s2 ) = 1, j2 (s2 ) = 1, j3 (s2 ) = 0,
 Musical Set Theories 23

j1 (s3 ) = 1, j2 (s3 ) = 1, j3 (s2 ) = 1, . . . , j1 (s6 ) = 0, j2 (s6 ) = 0 and

j3 (s6 ) = 1. Thus, the cycle index is the polynomial
1
P(S3 ,X) (t1 , t2 , t3 ) = (t31 + 3t1 t2 + 2t3 )
6
Let Y be the set Y = {0, 1}, X = {1, 2, 3} and G = S3 . The weight
function is defined on Y by w(0) = 1, w(1) = z. The sum of weights
of G-orbits is by Pólya’s theorem the following counting series

P(S3 ,X) (z) = 1 + z + z 2 + z 3

obtained from P (S3 , X) by substituting tk by 1 + z k . For the sym-

metric group Sn , we have the following result.
Proposition 1.2. The cycle index of the symmetric group SN of a
set X = {1, 2, . . . , N } of N elements is given by

  j k
1 tk
P(SN ,X) =
jk ! k
j k

where the sum is taken over all j = (j1 , j2 , . . . , jN ) verifying


N
kjk = N
k=1

In the chord counting problem of the N tone music, the set X is

identified with ZN and the group G is the cyclic CN , dihedral DN ,
symmetric SN or affine AN group. A pitch class set corresponds to
a characteristic function taking the pitches in the set to 1 and the
remaining notes to 0. The set of all pcsets is denoted by Y X with
X = ZN and Y = {0, 1}, or F (ZN ). The group G acts on F (X)
and induces an equivalence relation on k -chords. A pcset of length k
or a k -chord is a subset of the Cartesian product X k with cardinal
number k.
Definition 1.21. Let G be a group acting on the set F (ZN ) of
pcsets. The elements of the quotient set F (ZN )/G define the pcset
classes or simply collections relatively to the action of the group G.
A pcset class of cardinality k is called a collection of length k.
24 A Compendium of Musical Mathematics

Example 1.12. For the dihedral group DN , the elements of the

quotient set F (ZN )/DN are called the set classes under the action
of the dihedral group or simply the d-classes. For the cyclic group
CN , the elements of the quotient set F (ZN )/CN are called the set
classes under the action of the cyclic group or simply the c-classes
or musical collections.
The number of k -chord classes is the number of G-orbits which is
the coefficient of z k in the cycle index for variables

tk = 1 + z k

Proposition 1.3. The cycle index of the cyclic group Cn is the

polynomial
 
1 N d
P(CN ,ZN ) (t1 , . . . , tN ) = ϕ tN/d
N d
d|N

where ϕ is the Euler totient function.

Example 1.13. For N = 12, the cycle index is
1 12
P(C12 ,Z12 ) (t1 , . . . , t12 ) = (t + t62 + 2t43 + 2t34 + 2t26 + 4t12 )
12 1
Substituting tk by 1 + xk , we get the following counting series:

P(C12 ,Z12 ) (x) = 1 + x + 6x2 + 19x3 + 43x4 + 66x5 + 80x6

+66x7 + 43x8 + 19x9 + 6x10 + x11 + x12

Each coefficient of xk is the number of k -chord classes under the

CN group action. There are, for example, 19 trichords and 80 hexa-
chords.
Corollary 1.2. The number of k-chord classes under the CN group
action (up to transpositions) is given by the coefficients of the gen-
erator polynomial
 
1 N d
PCN (x) = ϕ 1 + xN/d (1.1)
N d
d|N
 Musical Set Theories 25

Definition 1.22. Euler’s totient function ϕ for integer m is the num-

ber of positive integers not greater than and coprime to m.
For m = p prime, ϕ(p) = p − 1 since all numbers less than p
are relatively prime to p. If m = pα is a power of a prime p, then
ϕ(p) = pα (1 − 1p ).
Proposition 1.4. If m1 and m2 are coprime, then totient function
has the multiplicative property:
ϕ(m1 m2 ) = ϕ(m1 )ϕ(m2 )
Example 1.14. The first values are as follows: ϕ(1) = 1, ϕ(2) = 1,
ϕ(3) = 2, ϕ(4) = 2, ϕ(5) = 4, ϕ(6) = 2, ϕ(7) = 6, ϕ(8) = 4, ϕ(9) = 6,
ϕ(10) = 4, ϕ(11) = 10, ϕ(12) = 4, etc.
Proposition 1.5. The cycle index of the dihedral group DN is given
by
⎧ −1)/2
⎨ 12 P(CN ,ZN ) + 12 t1 t(N
2 if N is odd
P(DN ,ZN ) =
⎩ 1 1 2 (N −2)/2 N/2
2 P(CN ,ZN ) + 4 (t1 t2 + t2 ) if N is even
Example 1.15. For N = 12, the cycle index is
1
P(D12 ,Z12 ) (t1 , . . . , t12 ) = (t12 + t62 + 2t43 + 2t34 + 2t26 + 4t12 )
24 1
1 1
+ t21 t52 + t62
4 4
k
Substituting tk by 1 + z , we get the following counting series:
P(D12 ,Z12 ) (z) = 1 + z + 6z 2 + 12z 3 + 29z 4 + 38z 5 + 50z 6
+ 38z 7 + 29z 8 + 12z 9 + 6z 10 + z 11 + z 12
Corollary 1.3. The number of k-chord classes under the DN group
action (up to transpositions and inversions) is given by the coeffi-
cients of the generator polynomial
⎧1
2 (N −1)/2
⎪ 1
⎪ 2 PCN (x) + 2 (1 + x)(1 + x ) if N is odd
⎪
⎨
PDN (x) = 12 PCN (x) + 14 ((1 + x2 (1 + x2 )(N −2)/2
⎪
⎪
⎪
⎩
+ (1 + x2 )N/2 ) if N is even
(1.2)
26 A Compendium of Musical Mathematics

For the affine group AN , the result is more complicated. Wan-Di

Wei and Ju-Yong (1993) have established the following results.

Proposition 1.6. The cycle index of the affine group AN is for

N = 2a

1
P(A2 ,Z2 ) (t1 , t2 ) = (t21 + t2 )
2
1
P(A4 ,Z4 ) (t1 , t2 , t3 , t4 ) = (t41 + 2t21 t2 + 3t22 + 2t4 )
8

and for a ≥ 3
⎛
a−1 

1 ⎝22(a−1)−1 t2a + a−i
P(A2a ,Z2a ) = 22(i−1) + ϕ(2i−1 )2a−1 t22i
22a−1
i=1
 2a−i−1⎞

a−2  a−i a−i−1 −1
i
+ ϕ(2i ) 2i t21 + 2a−1 t21 t22 t2k ⎠
i=0 k=1

Proposition 1.7. For a prime number p different from 2, the cycle

index P = P(Apa ,Zpa ) of the affine group An is for N = pa and a ≥ 1
⎛
1 a
a−i 
a−1 
P = ⎝ p2(i−1) (p − 1) tpi + pi+δ(d)(a−i)
p2a−1 (p − 1) p
i=1 i=0 d|p−1
 
pa−i−1 (p−1)/d
⎞
i
(pa−i−1 −1)/d ⎠
× ϕ(pi d) t1 td tp k d
k=1

where δ(x) = 1 if x > 1 and δ(1) = 0.

Proposition 1.8. Let X, Y be two finite sets and G, H be two finite

groups acting on X and Y. The cycle index of the natural action of
G × H on X × Y is the product

P(G×H,X×Y ) = P(G,X) ∗ P(H,Y )

 Musical Set Theories 27

The product of two polynomials

 N  m
P (t1 , t2 , . . . , tN ) = a(i) tikk , Q(t1 , t2 , . . . , tm ) = b(j) tjkk
(i) k=1 (j) k=1

where the sum is taken over n or m-tuples (j) = (j1 , . . . , jm ) is

defined by

 N m
i j gcd(k,l)
P ∗Q= a(i) b(j) k l
tlcm(k,l)
(i) (j) k=1l=1

where lcd(k, l) is the smallest integer which is a multiple of k and l,

and gcd(k, l) = kl / lcd(k, l). In particular, for N = pa11 · · · par r , where
pj are prime numbers, the cycle index of ZN is the product

r
P(AN ,ZN ) = ∗ P 
j=1 A aj ,Z aj
p p
j j

Example 1.16. For p = 3, the cycle index of A3 is

1 3
P(A3 ,Z3 ) (t1 , t2 , t3 ) = (t + 3t1 t2 + 2t3 )
6 1

The product leads to

P(A12 ,Z12 ) = P(A3 ,Z3 ) ∗ P(A4 ,Z4 )

1  12
= t + 3t41 t42 + 2t43 + 2t61 t32
48 1
+ 6t21 t52 + 4t23 t6 + 12t62 + 6t26 + 8t34 + 4t12

The substitution of tk = 1 + z k yields the enumeration series

P(A12 ,Z12 ) (z) = 1 + z + 5z 2 + 9z 3 + 21z 4 + 25z 5 + 34z 6

+ 25z 7 + 21z 8 + 9z 9 + 5z 10 + z 11 + z 12
28 A Compendium of Musical Mathematics

Summarizing the number of collections from each class, we get

the following table:
k-chords Mazzola (An ) Costère (Cn ) Forte (Dn ) Estrada (Sn )
1 Unison 1 1 1 1
2 Intervals 5 6 6 6
3 Trichords 9 19 12 12
4 Tetrachords 21 43 29 15
5 Pentachords 25 66 38 13
6 Hexachords 34 80 50 11
7 Heptachords 25 66 38 7
8 Octachords 21 43 29 5
9 Enneachords 9 19 12 3
10 Decachords 5 6 6 2
11 Endecachords 1 1 1 1
12 Dodecachords 1 1 1 1
Total 157 351 223 77

Of course, it is possible to consider other groups. For example, among

the five imprimitive groups of permutations of order 12, the group
generated by the three permutations which interchange the notes of
the augmented triad = (0, 4, 8) (1, 5, 9) (2, 6, 10) (3, 7, 11), of
the tritone = (0, 6) (1, 7) (2, 8) (3, 9) (4, 10) (5, 11) and of the
seventh diminished chord = (0, 3, 6, 9) (1, 8, 7, 2) (4, 11, 10, 5) has
a structure close to the cyclic group C12 . It has the same number of
k -chords except for the tetrachords for which it has 44 classes instead
of 43.

1.5 Exercises

Exercise 1.1
Let us study the sets of limited transposition for N = 6. By imitating
the proof of Theorem 1.4, find the four sets of limited transposition.

Exercise 1.2
In the N = 6 universe, using the results of the previous exercise,
explain the enumeration of chords as it is done in Section 1.4. Find
these results by computing the generator polynomial (1.1). Establish
the table of the number of k-chords up to transposition and the
number of sets of limited transposition.
 Musical Set Theories 29

Exercise 1.3
Let us study the sets of limited transposition for N = 8. By imitating
the proof of Theorem 1.4, find the five sets of limited transposition.
Exercise 1.4
In the N = 8 universe, using the results of the previous exercise,
explain the enumeration of chords as it is done on Section 1.4. Find
these results by computing the generator polynomial (1.1). Establish
the table of the number of k-chords up to transposition and the
number of sets of limited transposition.
Exercise 1.5
Write down the generator polynomial (1.2) for the action of the dihe-
dral group for N = 6.
Exercise 1.6
Write down the generator polynomial (1.2) for the action of the dihe-
dral group for N = 8.
Exercise 1.7
In the N = 8 universe, we consider the group T, I, M3  generated by
the transpositions, the inversions and the multiplication by 3 mod 8.
Show that the inverse of M3 is M5 the multiplication by 5 mod 8 and
write down the associated permutation. Compute the order of this
group.

References

Aigner, M. (2007). A Course in Enumeration, Springer, Berlin.

Clough, J. Engebretsen, N. and Kochavi, J. (1999). Scales, sets, and interval
cycles: A taxonomy, Music Theory Spectrum 21(1), 74–104.
Costère, E. (1954). Lois et styles des harmonies musicales, Presses universitaires
de France, Paris.
Costère, E. (1962). Mort ou transfiguration de l’harmonie, Presses universitaires
de France, Paris.
Estrada, J. (1994). Théorie de la composition, discontinuum-continuum, Thèse
de l’université de Strasbourg, sous la direction de François-Bernard Mâche.
Fripertinger, H. (1992). Enumeration in musical theory, Beiträge zur Elektronis-
chen Musik 1.
Hanson, H. (1946). Harmonic Materials of Modern Music: Resources of the Tem-
pered Scale, Appleton-Century-Crofts, New York.
Jedrzejewski, F. (2006). Mathematical Theory of Music, Ircam/Delatour, Paris.
30 A Compendium of Musical Mathematics

Loquin, A. (1874). Tableau de tous les effets harmoniques d’une à cinq notes
inclusivement, Féret, Bordeaux.
Loquin, A. (1884). Une révolution dans la science des accords, algèbre de
l’harmonie, traité complet d’harmonie moderne, sans notes de musiques ni
signes équivalents, avec des nombres pour représenter les effets harmoniques
et des lettres pour exprimer les mouvements de basse, Richault, Paris.
Mazzola, G. (1985). Gruppen und Kategorien in der Musik. Entwurf einer math-
ematischen Musiktheorie, Heldermann Verlag, Berlin.
Messiaen, O. (1944). Techniques de mon langage musical, Leduc, Paris.
Messiaen, O. (1996–2002). Traité du rythme, de couleur et d’ornithologie, Vol. 7,
Alphonse Leduc, Paris.
Morris, R. (1991). Class Notes for Atonal Music Theory, Frog Peak Music,
Hanover.
Parzysz, B. (1984). Musique et mathématique, Publication de l’Association des
professeurs de mathématiques de l’enseignement public, Paris.
Piňos, A. (1971). Tónové skupiny, Edition Supraphon, Prague. English transla-
tion, Tone Groups, Janáčkova akademie múzických uměnı́, Brno, 2001.
Starr, D. (1978). Sets, invariance and partitions, Journal of Music Theory 22(1),
1–42.
Wei, W.-D. and Xu, J.-Y. (1993). Cycle index of direct product of permutation
groups and number of equivalence classes of subsets of Zv , Discrete Math-
ematics 123, 179–188.
 Chapter 2

Generalized Interval Systems

Music theorist, David Lewin, in his seminal work, Generalized

Muscial Intervals and Transformations (Lewin, 1987) introduced a
new interval concept to apply to any musical space. He defined a
mathematical structure to model various musical space which allowed
him to establish more rigorous results. In this chapter, we review the
main concepts introduced by Lewin and his successors. Some results
can be established using the discrete Fourier transform. This area
has been treated by Emmanuel Amiot and we refer the reader to
his book (Amiot, 2016) and to the special issue of the Journal of
Mathematics and Music (Fiore, 2017).

2.1 Generalized Interval System

Definition 2.1. A Generalized Interval System (GIS) is a triple

(S, G, int) where S is a finite set of elements, G a group of inter-
vals, and int : S × S → G is a function from S × S to G such that
(1) int(x, y)int(y, z) = int(x, z) for all x, y, z ∈ S
(2) for every x ∈ S, and for every g ∈ G, there is a unique y ∈ S
such that int(x, y) = g.
If G is Abelian, we sometimes use the additive notation

int(x, y) + int(y, z) = int(x, z), ∀x, y, z ∈ S

31
32 A Compendium of Musical Mathematics

Example 2.1.
(1) The triple (Z, G, int) where G is the additive group of integers
(Z, +) is a GIS for the function int(x, y) = y − x.
(2) The triple (Z12 , G, int) where G is the additive group of integers
modulo 12 (Z12 , +) is a GIS for the function int(x, y) = y − x
mod 12.
(3) The triple (G, G, int) where S is a multiplicative group G is a
GIS for the function int(x, y) = yx−1 .
(4) If S is the diatonic scale isomorphic to Z7 , the triple
(Z7 , (Z, +), int) is a GIS for the function int(x, y) = y −x mod 7.
(5) In an Equal Division of the Octave with N degrees, the triple
(ZN , G, int) where G is the additive group of integers modulo
N (ZN , +) is a GIS for the function int(x, y) = y − x mod N.
(6) If G = D12 is the dihedral group of transpositions and inversions,
the triple (Z12 , D12 , int) is the space of Allen Forte’s set theory.
(7) If G is the permutation group on S, the triple (S, G, int) is a GIS
for the function int(x, y) = σ if and only if σ(x) = y.
(8) In order to modelize rhythms and durations, one can consider
the set of real numbers and its additive group (R, (R, +), int).
The function int(x, y) defines the number of time units between
x and y.
(9) If G = Q is the group of rational numbers and S is a set of
frequencies of the pitches, the triple (S, Q, int) is a GIS for the
function int : S × S → Q int(x, y) = x/y. This is the GIS of Just
intonation.
Definition 2.2. Let (S, G, int) be a GIS. A function f : S → S
preserves intervals if for all x, y ∈ S

int(f (x), f (y)) = int(x, y)

Definition 2.3. Let (S, G, int) be a GIS and n an element of G. The

function Tn : S → S is a transposition of interval n if for all x ∈ S

int(x, Tn (x)) = n

Proposition 2.1. For all i, j ∈ G, one has Ti Tj = Tji (with the

multiplicative notation for G).
 Generalized Interval Systems 33

Proof. Let x ∈ S, by the definition of the interval function

int(x, Ti Tj (x)) = int(x, Tj (x))int(Tj (x), Ti Tj (x))
= ji = int(x, Tij (x))
The proposition is true for all x, thus Ti Tj = Tji . 

Proposition 2.2. The function φ : i → Ti is bijective and φ(ji) =

φ(i)φ(j).
Proof. Let i, j be elements of G. Suppose that φ(i) = φ(j). By def-
inition of φ, one has Ti = Tj and thus for all x ∈ S, Ti (x) = Tj (x).
Therefore,
i = int(x, Ti (x)) = int(x, Tj (x)) = j
The function φ is thus injective. Since φ is surjective by construction,
one deduce that φ is bijective. The relation
φ(ji) = φ(i)φ(j)
is a consequence of the previous proposition. 
Theorem 2.1. Let (S, G, int) be a GIS. Denote by Y = {Ti }i∈G the
set of transpositions on S. Then Y is a permutation group and
Te = idS , Ti−1 = Ti−1
Proof.
(1) Let Te ∈ Y, e the neutral element of G, we have
int(x, Te (x)) = e = int(x, x)
Therefore, Te (x) = x and then Te is the identity on S.
(2) Let Ti ∈ Y. Since Ti−1 ∈ Y, the product Ti−1 Ti = Te = Ti Ti−1 ,
implies that Ti−1 = Ti−1 .
(3) We would like to show that Ti is a permutation of S, that is a
bijection from S to S. Let i ∈ G and x, y ∈ S. If Ti (x) = Ti (y),
then, composing by Ti−1 , one has Ti−1 Ti (x) = Ti−1 Ti (y), that is
x = y. Thus, Ti is injective. Furthermore, for x ∈ S, there exists
y = Ti−1 (x) ∈ S such that Ti Ti−1 (x) = x. Thus Ti is surjective
and then bijective.
34 A Compendium of Musical Mathematics

(4) Y is a group since the composition of transpositions is a binary

operation, associative and, following the previous properties,
having a neutral element e, and such that each element has an
inverse. This means that Y is a group.


Definition 2.4. Let (S, G, int) be a GIS and a, b ∈ S. The inversion

of a and b is the function Iab : S → S such that for all x ∈ S,

int(x, Iab (x)) = int(x, a)

Theorem 2.2. The inversion Iab is a bijection from S to S such that

Iab (b) = a and Iab (a) = b.
Proof. Let x, y ∈ S such that Iab (x) = Iab (y). By definition,

int(Iab (x), b) = int(Iab (y), b)

implies

int(a, x) = int(a, y)

and thus x = y, so Iab is injective. To show that the inversion is

subjective, consider a point y ∈ S, such that int(b, y) = j and a ∈ S.
By definition of a GIS, there exists a unique point x ∈ S such that
int(a, x) = j. Therefore,

int(b, y) = int(a, x) = int(b, Iab (x))

This implies that there exists a point x ∈ S such that Iab (x) = y.
Thus the function Iab is onto, and bijective since Iab is one-to-one and
onto. The previous equality,

int(b, Iab (b)) = int(b, a)

shows that Iab (b) = a. In the same way, we show that Iab (a) = b. 

Remark 2.1. If the group is Abelian, in additive notation, the inver-

sion Iab (x) is given by Iab (x) = −x + n for n = a + b.
 Generalized Interval Systems 35

2.2 Interval Function

Definition 2.5. Let X, Y be two finite sets of the GIS (S, G, int).
The interval function ifunc(X, Y ) of X and Y is defined by, for all
n ∈ G,
ifunc(X, Y )(n) = card{(x, y) ∈ X × Y, int(x, y) = n}

Example 2.2. Let X = {3, 7} and Y = {0, 4, 9} be two subsets of

Z12 . The interval function is defined by
ifunc(X, Y )(n) = card{(x, y) ∈ X × Y, y − x = n mod 12}
and the vector
ifunc(X, Y ) = (0, 1, 1, 0, 0, 1, 1, 0, 0, 2, 0, 0)
indicates that the semitone interval (n = 1) only appears once
(between 3 and 4) and the seventh (n = 9) appears twice (between
3 and 0, and between 7 and 4).
Theorem 2.3. Let (S, G, int) be a GIS, and X, Y two finite sets
of S. For all n ∈ G,
ifunc(Y, X)(n) = ifunc(X, Y )(n−1 )
Proof. The theorem results from the property
int(y, x) = n ⇔ int(x, y) = n−1

In the following, and unless otherwise indicated, we assume that
G is a Abelian group in additive notation. The interval function is
expressed using indicator functions. Let 1X be the indicator function
on X (1X (u) = 1 if u ∈ X and 1X (u) = 0 otherwise), and define the
adjunct of the indicator function by 1∗X (u) = 1X (−u).
Proposition 2.3. The interval function ifunc(X, Y ) of two finite
subsets X and Y of S is the convolution of 1∗X ∗ 1Y (i),
 
ifunc(X, Y )(n) = 1∗X (j)1Y (n − j) = 1X (k)1Y (n + k)
j k
36 A Compendium of Musical Mathematics

Theorem 2.4. Let (S, G, int) be a commutative GIS. The interval

function ifunc(X, Y ) of two finite subsets X and Y of S is
ifunc(X, Y )(n) = card(Tn (X) ∩ Y )
Proof. The proof is simply the rewriting in set terms of the previ-
ous equations
 
ifunc(X, Y )(n) = 1X (k)1Y (n + k) = 1X (j − n)1Y (j)
k j

= 1Tn (X) (j)1Y (j) = card(Tn (X) ∩ Y )
j


Theorem 2.5. The interval function ifunc(X, Y ) of two finite sub-

sets X and Y of S satisfies the following properties:
(1) ifunc(Ti (X), Y )(n) = ifunc(X, Y )(n + i)
(2) ifunc(X, Ti (Y ))(n) = ifunc(X, Y )(n − i)
(3) ifunc(Ti (X), Ti (Y ))(n) = ifunc(X, Y )(n)
Proof.
(1) By using the expression of the convolution, we get

ifunc(Ti (X), Y )(n) = 1Ti (X) (k)1Y (n + k)
k

= 1X (k − i)1Y (n + k)
k

= 1X (j)1Y (n + i + j)
j

= ifunc(X, Y )(n + i)
(2) In the same way, we have

ifunc(X, Ti (Y ))(n) = 1X (k)1Ti (Y ) (n + k)
k

= 1X (k)1Y (n + k − i)
k

= ifunc(X, Y )(n − i)
 Generalized Interval Systems 37

(3) For the third equation,


ifunc(Ti (X), Ti (Y ))(n) = 1Ti (X) (k)1Ti (Y ) (n + k)
k

= 1X (k − i)1Y (n + k − i)
k

= 1X (j)1Y (n + j) = ifunc(X, Y )(n)
j


Remark 2.2. If G is not Abelian, the result remains valid. In the

multiplicative notation,

ifunc(Ti (X), Ti (Y ))(n) = ifunc(X, Y )(ini−1 )

Definition 2.6. In the common set Z12 , the interval vector of a set
of pitch class set X is the vector whose coordinates are the values of
the interval function

iv(n) = ifunc(X, X)(n) = card(X ∩ Tn (X))

The first coordinate is the cardinal of X since ifunc(X, X)(0) =

card(X).
Definition 2.7. The interval-class vector icv(X) of a pitch-class set
X ⊂ Z12 is defined by:

icvX (n) = card{(x, y) ∈ X × X, x < y, and

(y − x = n or x − y = n)}

for n = 1, 2, . . . , 6.
Example 2.3. The interval vector of set X = {0, 4, 7} is iv(X) =
(3, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0, 0), and its interval-class vector is icv(X) =
(3, 0, 0, 1, 1, 1, 0). The interval vector of set X = {0, 4, 6} is
iv(X) = (3, 0, 1, 0, 1, 0, 2, 0, 1, 0, 1, 0) and its interval-class vector is
icv(X) = (3, 0, 1, 0, 1, 0, 1).
Theorem 2.6. Let X be a subset S = Z12 , and let iv(X) = (n0 ,
n1 , . . . , n11 ) be the interval vector of X and icv(X) = (m0 , m1 , . . . ,
m6 ) the interval-class vector, then
38 A Compendium of Musical Mathematics

(1) nj = mj for j = 0, 1, . . . , 5,
(2) n6 = 2m6
(3) nj = m12−j for j = 7, 8, . . . , 11.
In symbolic form, the interval function is made up of the number
of elements of X, the interval-class vector and its retrogradation.
−→ ←−
ifunc(X, X) = iv(X) = (card(X), icv(X), icv(X))
This result is generalized in the N -EDO.
Theorem 2.7. Let X be a subset S = ZN , and let iv(X) =
(n0 , n1 , . . . , nN −1 ) be the interval vector of X and icv(X) =
(m0 , m1 , . . . , mN/2 ) the interval-class vector, then
(1) nj = mj for j = 0, 1, . . . , N/2 − 1,
(2) nN/2 = 2mN/2
(3) nj = mN −j for j = N/2 + 1, . . . , N − 1.
where x is the ceiling function. In other words, if N = 2k + 1 is
odd
icv(X) = (n0 , n1 , . . . , nk−1 , nk )
and if N = 2k is even
icv(X) = (n0 , n1 , . . . , nk−1 , nk /2)

2.3 Injection Number

Definition 2.8. Let (S, G, int) be a GIS. The injection number

inj(X, Y )(f ) of X into Y for transformation f is the number of ele-
ments x ∈ X such that f (x) ∈ Y

inj(X, Y )(f ) = card{x ∈ X, f (x) ∈ Y } = 1(f (x)∈Y )
x∈X

Theorem 2.8. Let (S, G, int) be a GIS. If f is a bijection de S, then

inj(X, Y )(f ) = card (f (X) ∩ Y )
 Generalized Interval Systems 39

Proof. The proof follows from the definition of the injection num-
ber and the bijectivity of f ,

inj(X, Y )(f ) = 1(f (x)∈Y ) = card (f (X) ∩ Y )
x∈X


Theorem 2.9. If f is a bijection, then

inj(X, Y )(f ) = inj(Y, X)(f −1 )

Proof. Since f is a bijection,

   
f f −1 (Y ) ∩ X = f f −1 (Y ) ∩ f (X) = Y ∩ f (X)

Furthermore,
 
card f −1 (Y ) ∩ X = card (Y ∩ f (X))

Thus by the previous theorem

inj(Y, X)(f −1 ) = inj(X, Y )(f )


The relationship between the injection number and the interval
function is known as the common tone theorem.

Theorem 2.10 (Common tone theorem). For a translation Ti ,

the injection number is linked to the interval function by the relation

inj(X, Y )(Tn ) = ifunc(X, Y )(n)

Proof. Since f = Tn is bijective, and since we have

inj(X, Y )(f ) = card (f (X) ∩ Y )

it follows that

inj(X, Y )(Tn ) = card (Tn (X) ∩ Y )


40 A Compendium of Musical Mathematics

If we set X = Y and apply the common tone theorem to the

pitch-class space ZN , the number of pitch classes common to X and
its transpositions Ti (X) is spanned by members of X.
Corollary 2.1. In the ZN universe,
ifunc(X, X)(i) = card(X ∩ Ti (X))
Theorem 2.11. Let (S, G, int) be a GIS, and X, Y two finite subsets
of S. Let f, g be two functions S → S. Suppose that g is bijective.
Then
(1) inj(g(X), Y )(f ) = inj(X, Y )(f g)
(2) inj(X, g(Y ))(f ) = inj(X, Y )(g−1 f )
(3) inj(g(X), g(Y ))(f ) = inj(X, Y )(g−1 f g)
Proof. (1) By the definition of the injection number,
inj(g(X), Y )(f ) = card(f (g(X) ∩ Y ))
= card(f g(X) ∩ Y ) = inj(X, Y )(f g)
(2) Since g is bijective, we have
inj(X, g(Y ))(f ) = card(f (X) ∩ g(Y )) = card(gg−1 f (X) ∩ g(Y ))
 
= card(g g −1 f (X) ∩ Y ) = card(g−1 f (X) ∩ Y )
= inj(X, Y )(g −1 f )
(3) By applying the two previous properties, we have from (1)
inj(g(X), g(Y ))(f ) = inj(X, g(Y ))(f g)
and from (2)
inj(X, g(Y ))(f g) = inj(X, Y )(g−1 f g)


Theorem 2.12. In the common pitch-class set Z12 , if A is an hex-

achord |A| = |Ac |, then for all bijection f ,
inj(A, A)(f ) = inj(Ac , Ac )(f )
In particular (for translations Tn ),
ifunc(A, A) = ifunc(Ac , Ac )
 Generalized Interval Systems 41

Proof. This result is a simple consequence of the properties of

indicator functions.
 11

inj(A, A)(f ) = 1(f (x)∈A) = 1A (k).1f −1 (A) (k)
x∈A k=0
  
= (1 − 1Ac (k)) 1 − 1f −1 (Ac ) (k)
k
  
= 1− 1Ac (k) − 1f −1 (Ac ) (k)
k k k

+ 1Ac (k).1f −1 (Ac ) (k)
k

= 12 − 6 − 6 + 1(f (x)∈Ac )
x∈Ac

= inj(Ac , Ac )(f )

Theorem 2.13. Let (S, G, int) be a GIS, and X a subset of S. For
all bijection f , the injection number satisfies
inj(X, X)(f ) + inj(X, X c )(f ) = card(X)
In particular, in Z12 , if A is a hexachord (|A| = |Ac | = 6), then
ifunc(A, Ac ) = 6 − ifunc(A, A)
Proof. By definition, we get
 
inj(X, X)(f ) + inj(X, X c )(f ) = 1f (x)∈X + 1f (x)∈X c
x∈X x∈X

= 1 =card(X)
x∈X 
More generally, we have the following result:
Theorem 2.14. Let (S, G, int) be a GIS, and f a bijection on S.
We have the following properties for any subset X and Y of S:
(1) inj(X, Y c )(f ) = card(X) − inj(X, Y )(f )
(2) inj(X c , Y )(f ) = card(Y ) − inj(X, Y )(f )
(3) inj(X c , Y c )(f ) = card(Y c ) − card(X) + inj(X, Y )(f )
42 A Compendium of Musical Mathematics

Proof.
(1) For any subset X and Y, we have
 
inj(X, Y c )(f ) + inj(X, Y )(f ) = 1f (x)∈Y c + 1f (x)∈Y
x∈X x∈X

= 1 =card(X)
x∈X

(2) From (1), we have

inj(X c , Y )(f ) = inj(Y, X c )(f −1 ) = card(X) − inj(Y, X)(f −1 )

The result follows from the relation

inj(Y, X)(f −1 ) = inj(X, Y )(f )

(3) From (2), we have

inj(X c , Y c )(f ) = card(Y c ) − inj(X, Y c )(f )

and from (1),

= card(Y c ) − card(X) + inj(X, Y )(f )



Theorem 2.15. Under the same assumptions, if card(Y c ) =

card(X) then inj(X c , Y c )(f ) = inj(X, Y )(f ).

Proof. The proof results from the application of (3) in case

card(Y c ) = card(X). 

Theorem 2.16. Let A be a hexachord of Z12 , then

inj(Ac , Ac )(f ) = inj(A, A)(f )

Proof. If A is a hexachord of Z12 , card(A) = card(Ac ), and apply-

ing (3) leads to the result. 
 Generalized Interval Systems 43

2.4 Babbitt’s Hexachord Theorem

Theorem 2.17 (Babbitt’s hexachord theorem). Let (S, G,

int) be a GIS with S = Z12 . Complementary hexachords have the
same interval-class vector

icv(A) = icv(Ac )

Proof. From equation

card(A ∩ Tn (A)) = inj(A, A)(Tn )

= inj(Ac , Ac )(Tn )
= card(Ac ∩ Tn (Ac ))

it induces that the interval function are the same for all n

ifunc(A, A) = ifunc(Ac , Ac )

As are the interval vectors

iv(A) = iv(Ac )

It follows that the interval-class vectors are the same.

icv(A, A) = icv(Ac , Ac )

The following corollary generalizes Babbitt’s hexachord theorem
for a pitch-class set X of any cardinality and X ⊂ ZN .
Corollary 2.2. Let (S, G, int) be a GIS, and X a subset of S = ZN .
Let iv(X) = (n0 , n1 , . . . , nN −1 ) be the interval vector of X a subset
of any cardinality and set

m = card(X c ) − card(X)

then the interval vector of X c is

iv(X c ) = iv(X) + m = (n0 + m, n1 + m, . . . , nN −1 + m)

44 A Compendium of Musical Mathematics

Proof. From Theorem 2.14(3) with X = Y, we have for all j,

ifunc(X c , X c )(Tj ) = card(X c ) − card(X) + inj(X, X)(Tj )

= m + inj(X, X)(Tj )
= m + ifunc(X, X)(j)

thus

iv(X c ) = iv(X) + m


Corollary 2.3. Let (S, G, int) be a GIS, and X a subset of S = ZN .

Let icv(X) = (m0 , m1 , . . . , mN/2 ) be the interval-class vector of X
and set

m = card(X c ) − card(X)

then the interval-class vector of X c is

icv(X c ) = (m0 + m, m1 + m, . . . , m5 + m, mN/2 + m) if m is odd

and

icv(X c ) = (m0 +m, m1 +m, . . . , m5 +m, (mN/2 +m)/2) if m is even

Proof. Applying Theorem 2.7 to the previous theorem leads to the

result. 

2.5 Interval Sum

Definition 2.9. Let (S, G, int) be a GIS. The function sum : S×S →
G is defined for all x, y, z ∈ S, by

sum(x, y) = int(z, x)sum(z, y)

If G is Abelian, we suppose that the function sum satisfies

sum(y, x) = sum(x, y)

Proposition 2.4. Let (S, G, int) be an Abelian GIS, and x, y,

a, b ∈ S. sum(x, y) = sum(a, b) if and only if int(b, x) = int(y, a).
 Generalized Interval Systems 45

Proof. By definition

sum(x, y) = int(b, x)sum(b, y)

and

sum(a, b) = int(y, a)sum(y, b)

Since G is commutative sum(b, y) = sum(y, b), the result follows. 

Definition 2.10. Let (S, G, int) be a GIS, and X, Y two finite sub-
sets of S. The function interval sum sfunc(X, Y ) is defined by, for all
n∈G

sfunc(X, Y )(n) = card{(x, y) ∈ X × Y, sum(x, y) = n}

Theorem 2.18. In the common GIS (ZN , G, int) where G is a com-

mutative group in additive notation, the interval sum function is

sfunc(X, Y )(n) = card(In (X) ∩ Y )

where In (x) = −x + n mod N is the inversion modulo N .

Proof. The interval sum is written

 
sfunc(X, Y )(n) = 1∗X (j)1Y (n + j) = 1X (−j)1Y (n + j)
j j
 
= 1X (n − k)1Y (k) = 1In (X) (k)1Y (k)
j j

= card(In (X) ∩ Y )


Theorem 2.19. Let (S, G, int) be a commutative GIS. We have the

following properties

(1) sfunc(Ti (X), Y )(n) = sfunc(X, Y )(n − i)

(2) sfunc(X, Ti (Y ))(n) = sfunc(X, Y )(n − i)
(3) sfunc(Ti (X), Ti (Y ))(n) = sfunc(X, Y )(n − 2i)
46 A Compendium of Musical Mathematics

Proof.
(1) From the previous result,

sfunc(Ti (X), Y )(n) = 1∗Ti (X) (j)1Y (n + j)
j

= 1Ti (X) (−j)1Y (n + j)
j

= 1X (−j − i)1Y (n + j)
j

= 1X (−k)1Y (n − i + k)
k

= sfunc(X, Y )(n − i)

The other two propositions are demonstrated in the same way.



Theorem 2.20. Let (S, G, int) be a commutative GIS. Set Sab the
sum sum(a, b) and Iab inversion function for a and b in S. For two
finite subsets X, Y ⊂ S, we have

sfunc(X, Y )(Sab ) = inj(X, Y )(Iab )

Proof. By definition of the interval sum and the previous results,

sfunc(X, Y )(Sab ) = card{(x, y) ∈ X × Y, sum(x, y) = sum(a, b)}

= card{(x, y) ∈ X × Y, int(b, x) = sum(y, a)}
= card{(x, y) ∈ X × Y, y = Iab (x)}
= card(Iab (X) ∩ Y )
= inj(X, Y )(Iab )


Remark 2.3. In the common EDO ZN , the inversion Iab is the usual
inversion In (x) for n = a + b = sum(a, b). In this particular case, we
find the equality

sfunc(X, Y )(n) = card(In (X) ∩ Y )

 Generalized Interval Systems 47

2.6 Indicator Function

David Lewin (2001) introduced the Fourier transform of a pitch-class

set from its indicator function.

Definition 2.11. Let X be a subset of ZN , the Discrete Fourier

Transform of the set X is the Fourier Transform of its indicator
function
2iπkx
 −
FX (x) = F (1X ) (x) = e N
k∈X

Example 2.4. Let X = {0, 3, 6, 9} ⊂ Z12 . Its Fourier Transform is

the function Z12 → C such that

3

2iπ3kx 
− 4 if x ∈ {0, 4, 8}
FX (x) = e 12 =
0 otherwise
k=0

We will not detail here the results on the Discrete Fourier Trans-
form which are too numerous. We refer the reader to the book of
Emmanuel Amiot (2016) and to the special issue of the Journal of
Mathematics and Music (Fiore, 2017).
The elementary properties of the discrete Fourier transform are
as follows:

Theorem 2.21. Let X be a subset of ZN , a ∈ ZN , and λ an invert-

ible element of Z×N = ZN \{0}. The Fourier transform satisfies the
following properties:

(1) FX+a (x) = e−2iπax/N FX (x),

(2) F−X (x) = FX (x),
(3) FλX (x) = FX (λx).

In particular, the module |FX | of the Fourier transform is invari-

ant by transposition, inversion and multiplication by an invertible
element of X.
Since the interval function of two sets X and Y is the convolu-
tion product of the indicator functions of these sets and the Fourier
48 A Compendium of Musical Mathematics

transform transforms a convolution product into a simple product,

it follows that
Theorem 2.22. The Fourier transform of the interval function is
the multiplication of the transforms of the indicator functions:
 
Fifunc(X, Y ) = F 1X F (1Y ) = FXFY
Therefore,
Theorem 2.23. The Fourier transform of the interval vector of X ⊂
ZN is the square of the Fourier transform
F (iv(X)) = |FX|2
The Fourier transforms of X and its complement are linked by
the relation
∀x ∈ Z×
N, F(ZN \X)(x) = −F(X)(x)
Since the value of the Fourier Transform at x = 0 is the cardinality
of the set X, the preceding equality is true for any value of x when X
and X c have the same cardinality. Under this hypothesis |X| = |X c |,
we obtain an elementary proof of Babbitt’s hexachord theorem. The
Fourier transform of the interval vector iv(X c ) of the complement of
X is equal to the Fourier transform of iv(X):
F (iv(X c )) = |FX c |2 = |−FX|2 = |FX|2 = F (iv(X))
By taking the inverse Fourier transforms, we get the hexachord
theorem
iv(X c ) = iv(X)
if X and X c have the same cardinality.

2.7 Homometric Sets

Homometricity is a generalization of Allen Forte’s Z-relation.

Definition 2.12. Two sets A and B of ZN are Z-related if they have
the same interval content:
(A ZN B) ⇔ ic(A) = ic(B)
 Generalized Interval Systems 49

In other words, A and B share the same interval vector. Since

transposition or inversion does not change the interval content, we
get a lot of trivially Z-related sets. To avoid trivial cases, we con-
sider set classes up to transposition and inversion, that is, from a
mathematical point of view, sets up to the action of the dihedral
group.
Definition 2.13. Let A ⊂ ZN , the mask polynomial is defined by

A(x) = xa ∈ Z[x]/(xN − 1)
a∈A

and the Fourier transform of A is defined by the Fourier transform

of its characteristic function:

∀n ∈ ZN ,  −n
1A (x) = A(ωN )

where ωN = e2iπ/N is the N -th root of unity.

Definition 2.14. The reflection of A is the polynomial whose expo-
nents are the inversion I(A) of A mod N

A∗ (x) = xb
b∈I(A)

I denotes the inversion I(x) = −x mod N .

Definition 2.15. The Patterson function of the set A is the auto-
correlation function given by the convolution product

P (x) = A ∗ A∗ (x) = A(x)A(x−1 ) mod (xN − 1)

The Patterson function is an equivalent of the interval function,

since we have the following result.
Theorem 2.24. The coefficients of the Patterson function are equal
to the components of the interval function:

 
N −1
A ∗ A∗ (x) = ci xi = iv(A)(i)xi
ci ∈ifunc(A,A) i=0
50 A Compendium of Musical Mathematics

Proof. Let A = {a1 , a2 , . . . , an }. We get

 n ⎛ n ⎞
 
A(x)A(x−1 ) = xai ⎝ x−aj ⎠
i=1 j=1
⎛ ⎞

n 
n
= ⎝xai x−aj ⎠
i=1 j=1


n
= xai −aj = ΔA(x)
i,j=1


Example 2.5. For N = 12 and A = {0, 2, 3, 5}, it is easy to verify

this result. Computing the ifunc values leads to

iv(A) = ifunc(A) = (4, 1, 2, 2, 0, 1, 0, 1, 0, 2, 2, 1)

and the Patterson function is

P (x) = (1 + x2 + x3 + x5 )(1 + x10 + x9 + x7 )

= 4 + x + 2x2 + 2x3 + x5 + x7 + 2x9 + 2x10 + x11

whose coefficients are exactly the values of the interval function.

The previous result leads to a new characterisation of homometric

sets.

Lemma 2.1. Two sets A and B of ZN are homometric if and only

if they have the same Patterson function.

In others words, if A(x) is the mask polynomial of a set A ⊂ ZN ,

then the mask polynomial of the distance set ΔA is the Patterson
function, namely

ΔA(x) = A(x)A(x−1 )

Example 2.6. The first well-known example of two homometric

sets was given by Lino Patterson in 1944: {0, 3, 4, 5} Z8 {0, 4, 5, 7}.
 Generalized Interval Systems 51

For N = 8 and A = {0, 3, 4, 5}, the Patterson function is

PA (x) = (1 + x3 + x4 + x5 )(1 + x5 + x4 + x3 )
= 4 + 2x + x2 + 2x3 + 2x4 + 2x5 + x6 + 2x7
and for B = {0, 4, 5, 7}, the Patterson function
PB (x) = (1 + x4 + x5 + x7 )(1 + x4 + x3 + x)
= PA (x)
is equal to the Patterson function of A. Thus the two sets are homo-
metric.
Lemma 2.2. Given two multisets U and V . Then U + V and U − V
are homometric.
Proof. It suffices to prove that Δ(U + V ) = Δ(U − V ). Let d be an
element of Δ(U + V ). This element is written as d = (u2 + v2 ) − (u1 +
v1 ) = (u2 − v1 ) − (u1 − v2 ) which belongs to Δ(U − V ). Conversely,
if e ∈ Δ(U − V ), then e = (u2 − v2 ) − (u1 − v1 ) which is equal to
e = (u2 + v1 ) − (u1 + v2 ) ∈ Δ(U + V ). 
Using the multiplicative notation, we get:
Lemma 2.3. Let A(x) = U (x)V (x) and B(x) = U (x)V (x−1 ), then
A and B are homometric sets.
Proof. Since
ΔA(x) = A(x)A(x−1 )
= U (x)V (x)U (x−1 )V (x−1 )
  
= U (x)V (x−1 ) U (x−1 )V (x)
= B(x)B(x−1 ) = ΔB(x)
the two sets are homometric. 
The main result has been established by Rosenblatt and Seymour
(Rosenblatt and Seymour, 1982).
Theorem 2.25. Two sets A and B are homometric if and only of
there exist polynomials U (x) and V (x) and an integer a such that
A(x) = U (x)V (x) and B(x) = ±xa U (x)V (x−1 ).
52 A Compendium of Musical Mathematics

Proof. By computing the distance mask polynomial, we have

ΔB(x) = xa U (x)V (x−1 )x−a U (x−1 )V (x)
= U (x)V (x)U (x−1 )V (x−1 )
= ΔA(x)
The equality of their Patterson function shows that the two sets
A and B are homometric. Conversely, suppose that A and B are
homometric sets. Consider the polynomial P (x) = gcd(A(x), B(x)).
Then we can find two polynomials QA and QB relatively prime such
that A(x) = P (x)QA (x) and B(x) = P (x)QB (x) and two other
polynomials RA and RB such that
QA (x) = V (x)RA (x) and QB (x−1 ) = V (x)RB (x)
where V (x) = gcd(QA (x), QB (x−1 )). Since A and B are homometric,
we have A(x)A(x−1 ) = B(x)B(x−1 ), which leads to
RA (x)RA (x−1 ) = RB (x)RB (x−1 )
By construction, RA and RB are relatively prime. Furthermore, RA
and RB ∗ are also relatively prime, otherwise this would contradict the

fact that QA and QB are relatively prime. Since RA (x) and RA (x−1 )
are relatively prime to both RB (x) and RB (x−1 ), RA (x) = ±xα and
RB (x) = ±xβ . Therefore, A(x) = ±xα P (x)V (x) and B(x) = ±
xβ P (x)V (x−1 ). Denote U (x) = ±xα P (x) and set a = β − α, we get
A(x) = U (x)V (x) and B(x) = ±xa U (x)V (x−1 ). 
For difference sets (see Chapter 8), we have the following
characterization.
Theorem 2.26. Let A be a k-subset of ZN . Then A is a (N, k, λ)-
difference set if and only if its Patterson function is in Z [ZN ]:
A(x)A(x−1 ) = k + λ(T (x) − 1) mod xN − 1
where T is the polynomial
T (x) = 1 + x + x2 + · · · + xN −1
Another way to approach homometry is to introduce the notion
of spectral unit as it was done by Rosenblatt (1984).
Definition 2.16. A unit S(x) of the algebra group Q[ZN ] is called
a spectral unit if S(x)S(x−1 ) = 1 mod (xN − 1).
 Generalized Interval Systems 53

Rosenblatt proved the following:

Theorem 2.27. Two sets A and B are homometric if and only if
there exists a spectral unit S ∈ Q[ZN ] such that

S(x)B(x) = A(x) mod (xN − 1)

Example 2.7. Let N = 8, A = {0, 3, 4, 5} and B = {0, 1, 3, 4}.

The polynomial S(x) = (1 + x2 − x4 + x6 )/2 is a spectral unit since
S(x)S(x−1 ) = 1. Moreover, S(x)B(x) = xA(x) which means that
B and A = T1 (A) = {1, 2, 3, 6} are homometric. Since A(−1) =
B(−1) = 0, the solution is not unique. The polynomial
1
S(x) = (−3 + x + x2 + x3 + x4 + x5 + x6 + x7 )
4
is a spectral unit such that S(x)B(x) = x5 A(x). Thus the sets B and
T5 (A) are homometric, and therefore, A and B are homometric.
Furthermore, Rosenblatt (1984) used these results to find all non-
trivially homometric sets of k = 4 notes.
Theorem 2.28. If A and B are two nontrivially homometric sets of
4 elements, then A and B are one of the following types: for N =
4(r + s),
 
N N N N N N
A = 0, , + s, +s+r and B = 0, − r, , +s
2 2 2 2 2 2
or if N = 13n is a multiple of 13,

A = {0, n, 4n, 6n} and B = {0, 2n, 3n, 7n}

However, it is unlikely that such results can be obtained for a

number of higher elements card(A) > 4.
Example 2.8. In the classical system of 12 tones, A. Forte has shown
that there is only one homometric pair of 4 notes {0, 1, 4, 6} and
{0, 1, 3, 7}, three pairs of 5 notes {0, 1, 2, 4, 7} and {0, 1, 3, 5, 6},
{0, 1, 2, 5, 8} and {0, 1, 3, 8, 9}, {0, 1, 2, 5, 9} and {0, 1, 3, 4, 8},
and 15 pairs of 6 notes, such as 6-Z29 {0, 1, 3, 6, 7, 10} and 6Z-50
{0, 1, 3, 6, 8, 9}. A systematic study of the first values of N shows
that there is only one 4-notes homometric pair in Z8 ({0, 1, 2, 5},
54 A Compendium of Musical Mathematics

{0, 1, 3, 4}) and three 5-notes homometric pairs in Z10 . In the Z13
universe studied by Erv Wilson and W. Sethares, there is a unique
4-notes pair ({0, 1, 3, 9}, {0, 1, 4, 6}) and only two 5-notes pairs
({0, 1, 2, 3, 6, 10}, {0, 1, 2, 4, 5, 8}) and ({0, 1, 2, 4, 7, 9}, {0, 1, 3,
5, 7, 8}).

Example 2.9. From N ≥ 16 elements, David Lewin noticed that

multiplets appear. But there are no multiplets for N = 17 and
N = 19. In his article, On extended Z-Triples (Lewin, 1981), he gave
the three triples of Z-associated hexachords mod 16 and the two
triples of Z-associated heptachords. But he does not give the three
quadruples of Z-associated octachords, which exhaustively complete
the list of homometric multiplets mod 16, namely:

{0, 1, 2, 3, 4, 7, 9, 12}, {0, 1, 2, 3, 5, 8, 9, 14}, {0, 1, 2, 3, 5,

10, 12, 13}, {0, 1, 2, 4, 5, 7, 9, 10}
{0, 1, 2, 3, 5, 8, 11, 12}, {0, 1, 2, 3, 6, 8, 9, 13}, {0, 1, 2, 4, 5,
10, 11, 13}, {0, 1, 2, 5, 6, 8, 9, 11}
{0, 1, 2, 4, 5, 8, 10, 12}, {0, 1, 2, 4, 6, 8, 9, 12}, {0, 1, 2, 4, 6, 9,
10, 14}, {0, 1, 2, 4, 8, 10, 12, 13}

2.8 Exercises

Exercise 2.1
Let’s consider the quarter-tone space, which we present in relation
to the usual notes Z24 = Z12 × Z2 . Each note of Z24 is coded (x, n),
where x is a point of Z12 (a note of our classical tempered space) and
n = 0 or 1. If n = 0, the note x is not altered and if n = 1, the note x
is raised by a quarter tone. In this system, (0,0) is C, (0,1) is C+1/4
tone, (1,0) is C , (1,1) is C sharp raised by a quarter tone (C+3/4
tone), (2,0) is D, etc. We also consider the additive group Z24 . Define
an interval function int(x, y) on Z12 ×Z2 such that (Z12 ×Z2 , Z24 , int)
is a GIS.

Exercise 2.2
In the N = 8 universe, we introduce M3 the multiplication by 3
mod 8. Let us consider the injection number inj(X, Y ) of two pcsets
 Generalized Interval Systems 55

X and Y

inj(X, Y )(M3 ) = card(M3 (X) ∩ Y )

Show that

inj(X, Y )(M3 ) = inj(Y, X)(M5 )

and

inj(M3 (X), Y )(M3 ) = inj(X, M5 (Y ))(M5 ) = card(X ∩ Y )

Exercise 2.3
In the N = 12 universe, we introduce M5 the multiplication by
5 mod 12. Let us consider the injection number inj(X, Y ) of two
pcsets X and Y

inj(X, Y )(M5 ) = card(M5 (X) ∩ Y )

Show that

inj(X, Y )(M5 ) = inj(Y, X)(M7 )

and

inj(M5 (X), Y )(M7 ) = inj(X, M7 (Y ))(M7 ) = card(X ∩ Y )

Exercise 2.4
In the N = 12 universe, we introduce the function ibang defined by

ibang(X, Y )(n) = card {(x, y) ∈ X × Y, y − 5x = n mod 12}

(1) Show that

ibang(X, Y )(n) = card (Tn M5 (X) ∩ Y )

(2) Let X = {0, 1, 4} and Y = {0, 2}, compute the vector

ibang(X, Y ).
(3) Show the following properties:

ibang(Ti (X), Y )(n) = ibang(X, Y )(n + 5i)

ibang(X, Ti (Y ))(n) = ibang(X, Y )(n − 5i)
ibang(Ti (X), Ti (Y ))(n) = ibang(X, Y )(n)
56 A Compendium of Musical Mathematics

Exercise 2.5
Show that the discrete Fourier transform of Tn (X) where X is a pcset
verifies

FTn (X) (x) = e−2iπnx/N FX (x)

Exercise 2.6
Show that X = {0, 1, 2, 5, 7} and Y = {0, 1, 3, 5, 6} are homomet-
ric sets in Z10 by calculating their interval vector. Compute their
Patterson polynomial and show using Theorem 2.26 that they are
(N, k, λ)-difference sets.
Exercise 2.7
Verify that the Z-relation is stable by multiplication. Let X be a set
in ZN , m an integer such that gcd(m, N ) = 1 and m = 1, N − 1.
Denote by Mm the multiplication modulo N , and if X and Y are
homometric sets, then Mm X and Mm Y are also homometric. For
instance, let N = 12, show that the sets X = {0, 1, 2, 3, 5, 6} and
Y = {0, 1, 2, 3, 4, 7} are homometric and compute their interval con-
tent. Verify that the sets M5 X and M5 Y are also homometric and
compute their interval content.

References

Amiot, E. (2016). Music Through Fourier Space: Discrete Fourier Transform in

Music Theory, Springer, Cham (Switzerland).
Fiore, T. (2017). Perfect balance and the discrete Fourier transform. Journal of
Music and Mathematics 17(2–3), 65–66.
Lewin, D. (1981). On extended Z-triples, Theory and Practice 7, 38–39.
Lewin, D. (1987). Generalized Musical Intervals and Transformations, Yale
University Press, New Haven.
Lewin, D. (2001). Special cases of the interval function between pitch-class sets
X and Y , Journal of Music Theory 45(1), 1–30.
Rosenblatt, J. and Seymour, P.D. (1982). The structure of homometric sets, SIAM
Journal on Algebraic Discrete Methods 3(3), 343–350.
Rosenblatt, J. (1984). Phase retrieval, Communications in Mathematical Physics
95, 317–343.
 Chapter 3

Generalized Diatonic Scales

Is there an analog to the standard major or minor scales in a

given equal division of the octave with N notes rather than 12?
This chapter addresses this question on the basis of mathematical
and music-theoretical arguments set up by French composer Alain
Louvier under the name of modes of progressive transposition.
It explores cross-connections to related work by other authors,
especially with neo-Riemannian theories. In certain ways, this
chapter takes up the concepts given around the concept of diatonic-
ity in Jedrzejewski (2008) in a simpler form and considers what the
concept of tonality could be. As already mentioned in the previous
article, a historically and theoretically interesting source to this ques-
tion is Wyschnegradsky’s 24 Preludes opus 22 composed in 1916.
We study sets of progressive transposition in the light of Lewin’s
injection number and interval function and show their unicity.
Then, we review the properties of microdiatonic scales. In the last
section, we introduce a generalized concept of major keys based on
the existence of a generalized tritone. We also defined a general-
ized relative minor for particular generalized major scale. In most
cases, the results are based on the uniqueness of scales with certain
properties. As we wanted to distinguish major chords from minor
chords, all classes are considered only up to transposition (and not
up to inversion).

57
58 A Compendium of Musical Mathematics

3.1 Sets of Progressive Transposition

Introduced by Alain Louvier (1997), sets of progressive transposition

are particular cases of Douthett’s P-Cycles (Douthett and Kranz,
2000). Louvier had a brilliant musical career. In 1968, he won the
last annual Prix de Rome for musical composition. He has been the
director of the Conservatoire de Paris (1986–1991), where he also
taught music analysis and orchestration (1991–2009). In his paper,
he explains how to build sets of progressive transposition (PT) in
the quarter and third tone universes. More generally, PT sets can
be defined in the N -equal division of the octave. The set X is a PT
set if the transposition of X at some level m has only one note of
difference with X.
Definition 3.1. Let 1 < m < N and gcd(m, N ) = 1. The set X is a
set of progressive transposition if X is not a set of limited transpo-
sition and
card(X ∩ Tm (X)) = card(X) − 1
This also means that there exist x, y ∈ ZN , such that Tm (X) =
(X\{x})∪{y}. In particular, if X is a PT set, then X has N different
transpositions.
Most of the properties of PT sets can be established using the
properties of the injection number and the interval function ifunc,
defined by David Lewin. Let (S, G, int) be a generalized interval
system. We have seen that the injection number of X into Y for
f denoted inj(X, Y )(f ) is the number of elements x of X such that
f (x) ∈ Y

inj(X, Y )(f ) = card{x ∈ X, f (x) ∈ Y } = 1(f (x)∈Y )
x∈X
It is easy to show that if f is a permutation of S, then
inj(X, Y )(f ) = card(f (X) ∩ Y )
and that
inj(X, Y )(f ) = inj(X, Y )(f −1 )
If f is the translation Tn (x) = x + n mod N , since Tn is a bijection,
the injection number is linked to the interval function
inj(X, Y )(Tn ) = card (Tn (X) ∩ Y ) = ifunc(X, Y )(n)
 Generalized Diatonic Scales 59

where the interval vector is defined for n ∈ ZN by the numbers,

iv(n) = ifunc(X, X)(n) = card(X ∩ Tn (X))
Proposition 3.1. X is a PT set if and only if its complement X c
is a PT set.
Proof. It follows from the property of the injection number that if
X is a PT set at level m,
card(X ∩ Tm (X)) = inj(X, X)(Tm ) = card(X) − 1
This implies that a single element x of X is mapped by Tm in the
complement of X. And since Tm is a bijection, card(X) − 1 elements
of the complement X c are mapped to itself and one element to X.
Thus, the complement of X is a PT set at level m. Conversely, the
same reasoning applies to the complement. Therefore, X is a PT set
at level m is an equivalence with X c is a PT set at level m, since the
complement of X c is X. 
A set of progressive transposition is invariant by transposition.
The simple example is the 12 major scales.
Proposition 3.2. X is a PT set if and only if all of its transposition
is also PT.
Proof. Let X be a PT set for index m. Then for all n ∈ ZN ,
card(Tn (X) ∩ Tm Tn (X)) = inj(Tn X, Tn X)(Tm )
= inj(X, X)(Tn−1 Tm Tn )
= inj(X, X)(Tm ) = card(X ∩ Tm (X))
thus Tn (X) is a PT set. 
Theorem 3.1. X is a PT set for index m if and only if
iv(X)(m) = card(X) − 1
Proof. We have
card(X ∩ Tm (X)) = inj(X, X)(Tm ) = ifunc(X, Y )(m) = iv(X)(m)
If X is a PT set, then card(X) − 1 = card(X ∩ Tm (X)) and thus
iv(X)(m) = card(X) − 1. Conversely, if iv(X)(m) = card(X) − 1,
then card(X ∩ Tm (X)) = iv(X)(m) = card(X) − 1 and thus X is a
PT set. 
60 A Compendium of Musical Mathematics

Example 3.1. In the 12-tone universe, for m = 7, the major scale

X = {0, 2, 4, 5, 7, 9, 11} is a PT set since it has iv(X)(7) = 6 =
card(X) − 1.

3.2 Well-Formed Scales

Let us recall some common definitions. Well-formed scales have been

introduced by Norman Carey and David Clampitt (1989).
Definition 3.2. A scale is said to be generated if it is obtained by an
iteration of Tn for some n. A generic interval is the number of scale
steps between notes in the scale. A specific interval is the clockwise
distance between pitch classes on the chromatic circle.
Example 3.2. The major scale {0, 2, 4, 5, 7, 9, 11} is generated by
the interval of a perfect fifth by repeated applications of T7 . For
every generic interval of the second, there are two specific intervals
of 1 semitone (minor second) or 2 semitones (major second).
Definition 3.3. A generated scale is well formed if each generating
interval spans the same number of steps.
Example 3.3. The pentatonic scale {0, 2, 5, 7, 9}, generated by per-
fect fifth {5, 0, 7, 2, 9}, is well formed since all perfect fifths span the
same number of steps (3):
5, 7, 9, 0, 2, 5, 7, 9, 0, 2, 5, 7, 9, 0, 2, 5
The major scale {0, 2, 4, 5, 7, 9, 11} is well formed since it is gen-
erated and all perfect fifths span the same number of steps.
The octatonic scale {0, 1, 3, 4, 6, 7, 9, 10} is not well formed since
it is not generated.
Definition 3.4. The chromatic size (specific interval) or the
c-distance between two notes a and b of a scale is the number of
steps between these two notes:
dc (a, b) = min(b − a mod N, a − b mod N )
Definition 3.5. The scale interval (generic interval) or the
d-distance between two notes a and b of a scale S is the number
of intervals or the number of notes between a and b in S:
dd (a, b) = #([a, b] ∩ S)
 Generalized Diatonic Scales 61

These notions lead us to another definition of well-formed scales.

Definition 3.6. A generated collection where a single d-distance
corresponds to the c-distance of the generator is called well formed.
The term maximally even (ME) was coined by John Clough and
Jack Douthett (1991) to refer to scales that are subsets of a chromatic
scale and in a well-defined sense are spread out as much as possible
within that chromatic.

Definition 3.7 (Clough and Douthett, 1991). A scale is maxi-

mally even if each scale intervals comes in either one or two chromatic
sizes of consecutive integers. In other words, a maximally even scale
has exactly two specific intervals for each generic interval.

Example 3.4. The octatonic scale {0, 1, 3, 4, 6, 7, 9, 10} and the

major scale {0, 2, 4, 5, 7, 9, 11} are maximally even. Many examples
of calculating distances can be found in Timothy Johnson’s book
(Johnson, 2008).

Scale intervals Chromatic sizes Scale intervals Chromatic sizes

d-distance c-distance d-distance c-distance
1 1, 2 1 1, 2
2 3 2 3, 4
3 4, 5 3 5, 6
4 6 4 6, 7

Jack Douthett’s set is defined as follows.

Definition 3.8. For N, k, m ∈ Z, with N > k and k = 0, the
J-function on Z is
 
m Nx + m
JN,k (x) =
k

where x is the floor function (the greatest integer less than or equal
to x) and Douthett’s J-set is
m
JN,k = {JN,k
m
(x), x = 0, 1, . . . , k − 1}

Theorem 3.2 (Clough and Douthett, 1991). Let N and d be

the chromatic and diatonic cardinalities, respectively, and let n be
62 A Compendium of Musical Mathematics

any fixed integer such that 0 ≤ n ≤ N − 1. The pcset JN,d n of car-

dinality d, where x is the greatest integer less than or equal to x
(floor function),
  
Nk + n
n
JN,d = , k = 0, . . . , d − 1
d
is maximally even.
We have a characterization of the subclass of ME(N, k) sets, with
N > k and gcd(N, k) = 1. That is the ME sets that are not degen-
erated well-formed.
Theorem 3.3. Let X be a subset of ZN with cardinality k prime to N.
X is ME set that is also non-degenerate well-formed, if and only
if X is a collection of successive images of some element x ∈ ZN
by the translation Tm , where n is the multiplicative inverse of k,
nk = 1 mod N :

X = {x, Tn (x), . . . , Tnk−1 (x)}

Remark 3.1. ME sets are invariant under the action of the T /I

group. The sets X and

X = {x, T−n (x), . . . , T−n

k−1
(x)}

obtained for −n belong to the same class.

Example 3.5. The octatonic set, determined with J by N = 12,
k = 8, is a degenerated well-formed ME set, since k is not coprime
with N .
Using the Fourier transform of a pcset A ⊂ ZN which is the
Fourier transform of its characteristic function
 2iπkt
FA (t) = e− N
k∈A

we have another characterisation of ME sets.

Theorem 3.4. The pcset A ⊂ ZN , with cardinality r, is a maximally
even set if the number |FA (r)| is maximal among the values |FX (r)|
for all pcsets X of cardinality r.
 Generalized Diatonic Scales 63

Definition 3.9. A scale S has the Myhill property if each scale inter-
val comes in two chromatic sizes. In other terms, S has exactly two
c-distances for every d-distance.
Example 3.6. The major scale {0, 2, 4, 5, 7, 9, 11} satisfies the
Myhill property.
Definition 3.10 (Clough et al., 1993). A P-cycle is a cycle of
three or more pairwise distinct sets (scales or chords) from the same
set class (sets equivalent under transposition and inversion) such that
there is a map between each pair of adjacent sets that leaves all but
one of its notes fixed. If the sets in the cycle exhaust the set class,
then the cycle is said to be unidirectional, otherwise it is a toggling
P-cycle.
All P-cycles are unidirectional for our purposes.
Example 3.7. The cycle of fifths is unidirectional
7T 7 T 7 T 7 T 7 T
{0, 7} −→ {7, 2} −→ {2, 9} −→ · · · −→ {5, 0} −→ {0, 7}

The following 6-cycle of minor and major chords is toggling

(e, E, g, G, c, C)
I
3 7 I 11 I
{4, 7, 11} −→ {4, 8, 11} −→ {8, 11, 3} −→ {8, 0, 3}
I
3 7 I 8 I
−→ {0, 3, 7} −→ {0, 4, 7} −→ {4, 7, 11}

Definition 3.11. In diatonic set theory, it is said that cardinality

equals variety when the number of notes in a subset X of a particular
scale S is equal to the number of distinct interval patterns formed
by diatonic transpositions of X on each note of S.
Example 3.8. Consider the subset X = {0, 2, 4} of the C major
scale S = {0, 2, 4, 5, 7, 9, 11}. Each diatonic transposition of X,
namely {0, 2, 4}, {2, 4, 5}, {4, 5, 7}, {5, 7, 9}, {7, 9, 11}, {9, 11, 0}
and {11, 0, 2}, involves three interval patterns: M2-M2, M2-m2 and
m2-M2. Three distinct pitch classes generate three distinct patterns:
cardinality (of X) equals variety (of diatonic transpositions).
The same concepts are transposed by counting no longer on the
chromatic circle but on the circle of fifths.
64 A Compendium of Musical Mathematics

Definition 3.12. In diatonic set theory, it is said that structure

implies multiplicity when the intervals between notes formed by the
shortest distance around the diatonic circle of fifths (structure) are
equal to the number of times that interval patterns appear in the
group of transposed series (multiplicity).

Example 3.9. Consider, as previously, the subset X = {0, 2, 4} of

the C major scale S = {0, 2, 4, 5, 7, 9, 11}. On the circle of fifths
5, 0, 7, 2, 9, 4, 11, (5), the distance between elements of X forms
the pattern 2-2-3, which is exactly the pattern formed by the trans-
posed sets of X: M2-m2 occurs twice ({2, 4, 5} and {9, 11, 0}), m2-M2
occurs twice ({4, 5, 7} and {11, 0, 2}) and M2-M2 occurs three times
({0, 2, 4}, {4, 5, 9} and {7, 9, 11}).

Theorem 3.5. Cardinality equals variety and structure implies mul-

tiplicity are true for all maximally even sets or sets with Myhill’s
property.

Definition 3.13 (Louvier). A set of progressive transposition is a

cycle of three or more pairwise distinct sets from the same set class
(sets equivalent under transposition only) such that there is a map
between each pair of adjacent sets that leaves all but one of its notes
fixed that exhaust the set class.

For each choice of index m, there exists only one class of set
of progressive transposition. More precisely, we have the following
result.

Theorem 3.6. For each generator 1 < m < N − 1 such that

gcd(m, N ) = 1, there exist a unique class set X (up to transposi-
tion) of progressive transposition given by

X = {x, Tm (x), . . . , Tm
k−1
(x)}

for some element x ∈ ZN and such that

(1) X has k elements,
(2) X has N different transpositions,
(3) X and Tm (X) have k − 1 notes in common.
Moreover, if k = m, X is well formed and maximally even.
 Generalized Diatonic Scales 65

Remark that if k = m, the set X is not necessary well-formed

(WF). For instance, N = 12, m = 7 and k = 4, X = {0, 2, 7, 9} is
not WF. X is not always maximally even (ME). For instance, for
N = 12, m = 7 and k = 3, X = {0, 2, 7} is not ME.
Example 3.10. Let N = 12. The integers of ZN coprime with N
are {1, 5, 7, 11}. Thus, 1 < m < 11 takes two values: m = 5 or m =
7 = −5. If m = 7, for x = 5, the set X = {0, 2, 4, 5, 7, 9, 11} = J12,7 5

corresponds to the white keys of the keyboard (the C major scale)

and its complement X c = {1, 3, 6, 8, 10} = T1 (J12,5 1 ) corresponds to

the black keys. If m = 5, the situation is dual. For x = 9, the set

Y = {9, T5 (9), . . . , T54 (9)} = {0, 2, 5, 7, 9} = T11 (X c ) corresponds to
the black keys and its complement corresponds to the white keys.
Proof. Since m is coprime to N , the set X = {x, Tm (x), . . . ,
Tmk−1 (x)} has N different transpositions: X is not a set of limited

transposition. Moreover, since the set

Tm (X) = {Tm (x), . . . , Tm

k
(x)}

has with X the intersection

X ∩ Tm (X) = {Tm (x), . . . , Tm

k−1
(x)}

the cardinality #(X ∩ Tm (X)) = k − 1 shows that X is PT. If k = m,

X is a well-formed scale and ME by the previous theorem. 

3.3 Generalized Diatonic Scales

Models of diatonicity have been proposed by many authors such as

Balzano (1980), Agmon (1989, 2013) or Broué (2001). The model we
propose in this section extends the model proposed in Jedrzejewski
(2008).
In the following, we introduce the multiplicative inverse r ∈ Z×
N
of m such that rm = 1 mod N , u the integer of ZN such that

u = max(−r, r) (3.1)

and v the integer of ZN given by

v = min{max(r, −r), max(r, 2r), max(−r, −2r)} (3.2)

66 A Compendium of Musical Mathematics

Theorem 3.7. Let 1 < m < N − 1 and gcd(m, N ) = 1, X a PT

scale of cardinal k and generator m. X does not have two consecutive
notes if and only if k ≤ min(−r, r).
Proof. Let jm be an element of the set X for j = 0, 1, 2, . . . , k − 1.
For r as above, the indexes {j, j + r} lead to two consecutive notes
{jm, jm + 1} since rm = 1 mod N . The indexes {j, j − r} also lead
to two consecutive notes. Thus, k ≤ min(−r, r). 

Corollary 3.1. Let 1 < m < N − 1 and gcd(m, N ) = 1, X a PT

scale of cardinal k and generator m. The complement of X does not
have two consecutive notes if and only if k ≥ u.
Theorem 3.8. Let 1 < m < N −1 and gcd(m, N ) = 1, X a PT scale
of cardinal k and generator m. X does not have three consecutive
notes if and only if k ≤ v.
Proof. Let jm be an element of the set X for j = 0, 1, 2, . . . , k − 1.
For r as above, the indexes {j, j + r, j + 2r} lead to three consecutive
notes {jm, jm + 1, jm + 2} since rm = 1 mod N . The indexes {j, j −
r, j + r} and {j, j − r, j − 2r} also lead to three consecutive notes. In
order to get all the notes in the same set, we need to consider
v = min{max(r, −r), max(r, 2r), max(−r, −2r)}
That is, if r > N/2, v = min(r, −2r) and if r < N/2, v = min(−r, 2r),
we get
⎧
⎪
⎪ 2r if 1 < r ≤ N/3
⎪
⎨ N −r if N/3 ≤ r < N/2
v=
⎪
⎪ r if N/2 < r ≤ 2N/3
⎪
⎩
2N − 2r if 2N/3 ≤ r < N

If N is prime and N > 2. The generator m takes ϕ(N ) = N − 1
values, where ϕ is the Euler totient function. The integers coprime
with N are m = 1, 2, 3, 4, . . . , N − 1. The first integer m greater that
N/2 is (N + 1)/2.
Proposition 3.3. If N is prime and N > 2, the set
X = {j(N + 1)/2, j = 0, 1, . . . , (N − 1)/2}
has three consecutive notes.
 Generalized Diatonic Scales 67

Proof. The three consecutive notes are (m, 3m, 5m) with m =
(N + 1)/2. Since 3(N + 1)/2 = (N + 1)/2 + N + 1 = (N + 1)/2 + 1
mod N , 5(N + 1)/2 = (N + 1)/2 + N + 2 = (N + 1)/2 + 2 mod N .


Proposition 3.4. If N is prime and N > 2, the set X = {j(N +3)/2,

j = 0, 1, . . . , (N + 1)/2} does not have three consecutive notes.
Proof. Let m = (N + 3)/2. The set X can be decomposed as
X = A ∪ B with

A = {3j, j = 0, 1, 2, . . . , (N + 1)/4}

and

B = {3j + m, j = 0, 1, 2, . . . , (N − 3)/4}

and thus X does not have three consecutive notes. 

If N is prime, the set X = {j(N + 3)/2, j = 0, 1, . . . , (N + 1)/2}
is the diatonic scale in the sense of the following section.
Definition 3.14. Let 1 < m < N − 1 such that 2m ≥ N + 1 and
gcd(m, N ) = 1, a set X is a diatonic scale of generator m if

(1) X has k elements, N different transpositions, and X and Tm (X)

have m − 1 notes in common,
(2) X does not have three consecutive notes (x, x + 1, x + 2) for some
x ∈ ZN ,
(3) X c does not have two consecutive notes (x, x + 1) for some
x ∈ ZN .

This definition means that X is a TP scale by (1). (2) means that

there is no three consecutive white keys, and (3) means that there is
no two consecutive black keys. The condition (2m − N ) ≥ 1 excludes
the dual cases.
Theorem 3.9. Let 1 < m < N − 1 such that 2m ≥ N + 1 and
gcd(m, N ) = 1, a set X is a diatonic scale of k elements and genera-
tor m if and only if X is a TP scale with k elements and generator m

X = {x, Tm (x), . . . , Tm
k−1
(x)}
68 A Compendium of Musical Mathematics

for some element x ∈ ZN and

u≤k≤v (3.3)

The set X is denoted by Diam

N.

For a given N and a given m as above, there is 0 or 1 diatonic

scale of generator m.

Example 3.11. For instance, for N = 13, the generator m belongs

to the set {7, 8, 9, 10, 11}. For each value m, the triplet (r, u, v)
given by Equations (3.1) and (3.2) leads to a set of values of k by
Equation (3.3).

Example 3.12. For m = 7, there is no diatonic scale since u > v in

the triplet (r, u, v) = (2, 11, 4).

Example 3.13. For m = 8, (r, u, v) = (5, 8, 8), there exists a dia-

tonic scale of k = 8 elements:

Dia813 = {0, 1, 3, 4, 6, 8, 9, 11}

and for m = 11, there exists also a diatonic scale of k = 7 elements:

Dia11
13 = {0, 1, 3, 5, 7, 9, 11}

The other values of m lead to a triplet with u > v.

From these scales, we can choose a unique diatonic class scale (up
to inversion and transposition) by choosing the smallest m.

Definition 3.15. The (optimal) diatonic scale DiaN of ZN is the

diatonic scale Diam
N obtained for the smallest generator m such that
2m ≥ N + 1.

Example 3.14. For N = 13, the smallest m is m = 8, Dia13 =

Dia813 . The smallest m = 7 is excluded since {0, 7, 1, 8, 2, 9, 3} is not
a 7-diatonic scale (it has three consecutive white keys). For N = 10,
there is no diatonic scale.

The optimal diatonic scale always exists as long as N ≥ 12.

 Generalized Diatonic Scales 69

Example 3.15. If N = 18, m ∈ {5, 7, 11, 13}. If m = 11, the 10 first

multiples 0, 11, 2, 13, 4, etc. form a set

X = {0, 1, 4, 5, 8, 9, 11, 12, 15, 16}

whose complement X c has two consecutive notes. If m = 13, the 11

first multiples form a set

Dia18 = {0, 1, 3, 4, 6, 8, 9, 11, 13, 14, 16}

which satisfies all the properties of the diatonic scale.

Example 3.16. If N = 20, m ∈ {3, 7, 9, 11, 13, 17}. If m = 11, the
eleven first multiples 0, 11, 2, 13, 4, etc. form a set

Dia20 = {0, 2, 4, 6, 8, 10, 11, 13, 15, 17, 19}

which has no three consecutive notes and its complement

Diac20 = {1, 3, 5, 7, 9, 12, 14, 16, 18}

has no two consecutive notes. If m = 17, the first 13 elements form

a diatonic scale

X  = {0, 2, 4, 5, 7, 8, 10, 11, 13, 14, 16, 17, 19}

which verifies the properties of the diatonic scale. But only Dia11
20
verifies the condition on m (the smallest integer such that 2m >
N + 1).
Proposition 3.5. The complement of X does not contain two con-
secutive notes if and only if X c ⊂ T1 (X).
Proof. If (x, x+1) ∈ X c , then T−1 (x, x+1) = (x−1, x) ∈ / X since x
is in X. Thus, (x, x+1) ∈ / T1 (X). Conversely, since the transpositions
are bijections, if (x, x+1) ∈ X, then T1 (x, x+1) = (x+1, x+2) ∈ / X c.


3.4 Generalized Major and Minor Scales

In the common equal temperament with 12 notes, the tritone plays an

important role in the modulations (changing keys). As a set of limited
70 A Compendium of Musical Mathematics

transposition, and as it belongs to the dominant seventh chord and

to the diminished seventh chord, the tritone articulates the change
from one key to another. Its unstable character sets it apart. The
unicity of the tritone in major scale characterizes its key.
Proposition 3.6. Let D be the set of divisors of N minus {1, N }.
For d ∈ D, and k such that N = dk, the set

Ld = {dj, j = 0, 1, . . . , k}

and its transposition Tx (Ld ) for all x ∈ ZN are sets of limited

transposition.
Example 3.17. For N = 12, D = {2, 3, 4, 6}. The set L6 = {0, 6}
and Tx (L6 ) = {x, x + 6} are sets of limited transposition. The sets
L4 = {0, 4, 8} and L3 = {0, 3, 6, 9} are also sets of limited transpo-
sition. If N is prime, the set D is empty and the N -EDO does not
have sets of limited transposition.
Proposition 3.7. Let N = dk. The set Ld has exactly k
transpositions.
The following theorem is the engine for changing keys.
Theorem 3.10. Let D be the set of divisors of N minus {1, N } and
X be a set with N transpositions. Suppose L ⊂ X, L = Tx (Lk ) for
some x, d ∈ D, and N = dk. Then L is included in exactly d sets of
class X, namely

{X, Tk (X), . . . , Tkd−1 (X)}

Proof. Since X = L∪(X\L) and L is a set of limited transposition,

Tkd (X) = Tkd (L) ∪ Tkd (X\L) = L ∪ Tkd (X\L)

thus L ⊂ Tkd (X). 

Definition 3.16. Let D be the set of divisors of N minus {1, N } and

d the greatest element of D, and q such that N = dq, the generalized
tritone is the class set (up to transpositions)

Ld = {dj, j = 0, 1, . . . , q − 1}
 Generalized Diatonic Scales 71

Proposition 3.8. Let 1 < m < N − 1 and gcd(m, N ) = 1, X a PT

scale of cardinal k and generator m. X contains a generalized tritone
if and only if k ≥ d(q − 1) + 1, where d is the greatest element of D
and N = dq.
Proposition 3.9. Let X be a diatonic scale with generator m in the
N universe. X contains at least one generalized tritone.
Definition 3.17. Let 1 < m < N − 1 such that 2m ≥ N + 1 and
gcd(m, N ) = 1, X is a generalized major (or micro-major) scale if
(1) X is a set of progressive transposition,
(2) X does not have three consecutive notes,
(3) X contains a unique generalized tritone.
Theorem 3.11. Let m be such that gcd(m, N ) = 1, r the solution
of rm = 1 mod N and v as above, X a PT set
X = {x, Tm (x), . . . , Tm
k−1
(x)}
of cardinality k. X is a generalized major scale if and only if
d(q − 1) + 1 ≤ k ≤ v (3.4)
A generalized major scale does not necessarily exist. If N is prime,
there is no generalized tritone and thus no major scale, although there
exists a diatonic scale.
Example 3.18. If N = 12, m ∈ {5, 7}. The tritone is the class set
{0, 6}. Since the parameters are r = v = 7, X does not have three
consecutive notes if and only if k ≤ 7 and X contains a transposition
of the tritone {0, 6} if and only if k > 6. Therefore, the PT set
X = {0, 2, 4, 5, 7, 9, 11} contains a unique tritone {5, 11}. In this case,
the major scale is the same as the diatonic scale.
Example 3.19. If N = 14, m ∈ {9, 11}. The generalized tri-
tones are transpositions of {0, 7}. If m = 9, r = 11, d = 7 and
v = min(r, −2r) = 6. X does not contains three consecutive notes if
and only if k ≤ 6 and X contains a transposition of the tritone {0, 7}
if and only if k > 7. Therefore, there is no major scale associated
with m = 9. The scale
X = {0, 3, 4, 7, 8, 9, 12, 13}
with m = 9, k = 8 is well-formed, PT and has a unique tritone but
has three consecutive notes.
72 A Compendium of Musical Mathematics

If m = 11, r = 9 and v = min(r, −2r) = 9. k needs to be less

or equal to 9 and greater than 7. So the candidates are k = 8 or 9.
The set formed by the nine first multiples of m which is the diatonic
scale

3 0 11 8 5 2 13 10 7

contains two generalized tritones and cannot be a generalized major

scale. Thus,

Dia14 = {0, 2, 4, 5, 7, 8, 10, 11, 13}

has nine elements. For k = 8, the first eight multiples define the
major scale with a unique tritone

Maj14 = {0, 2, 5, 7, 8, 10, 11, 13}

Although the scale Maj14 is a PT set, a subset of the diatonic scale

and has a unique tritone, Maj14 is neither maximally even nor well-
formed. But it has a unique tritone and no three consecutive notes,
which is an important criterion to keep the major character and avoid
the feeling of chromaticism. This example shows that the choice arises
between a well-formed set and TP set. Is it more important for a
major scale to be a well-formed scale or to be a TP set?
Example 3.20. If N = 15, m ∈ {2, 4, 7, 8, 11, 13}. The divisors of 15
are {1, 3, 5} so d = 5 and q = 3. The generalized tritone is {0, 5, 10}.
There exists a diatonic scale for m = 13

Dia15 = {0, 1, 3, 5, 7, 9, 11, 13}

but no major scale. A major scale needs to take 11 multiples of m,

thus has cardinality 11, but in this case, X has three consecutive
notes, which contradicts the definition.
Example 3.21. If N = 21, there exists a diatonic scale for m = 13
(or m = 8)

Dia21 = {0, 2, 4, 5, 7, 9, 10, 12, 13, 15, 17, 18, 20}

but no major scale for same reasons as for N = 15.

 Generalized Diatonic Scales 73

• Case N even
If N is even, we have a graphical representation of these scales.
The set {N − m, (N/2 − 1)m} is a generalized tritone. The major
scale with tonic 0 in the N universe is

MajN = {0, m, 2m, 3m, 4m, . . . , (N/2 − 1)m, N − m}

In this case, the diatonic scale is equal to the major scale. Its graphic
representation is in the frame (N − m, N/2) which has the multiples
of N − m along the x-axis and the multiples of N/2 along the y-axis.
A linear representation can be obtain if the two first notes {0, N −m}
are added after m on the right of the graph. The two ends of the
graph are then separated by the unique generalized tritone of the
major scale:

0 — N −m
|
(N/2 − 1)m — · · · — 4m — 3m — 2m — m

The relative minor scale has one more generalized tritone {2m, 2m +
N/2}

MinN = {0, 2m + N/2, 3m, 4m, 5m, . . . , (N/2 − 1)m, N − m}

The note m is replaced by 2m + N/2. The two scales are in the same
neighborhood (differ by only one note):

0 — N −m 2m + N/2
| |
(N/2 − 1)m — · · · — 4m — 3m — 2m

Example 3.22. For N = 12, m = 7. Maj12 is the C-major

scale, and Min12 is the A-minor scale (harmonic minor scale). For
N = 14, we find the previous results. The relative minor of the
major scale is the scale obtained from the major scale replacing
11 by 1, Min14 = {0, 1, 2, 3, 5, 8, 10, 13}. For N = 24, the diatoni-
cized chromatic scale was used by Ivan Wyschnegradsky. It is a scale
of 13 notes, constructed by connecting two heptachords (12222222-
12222222). The composer used this scale in his 24 Preludes op 22
74 A Compendium of Musical Mathematics

and Premier Fragment Symphonique, op. 23. Furthermore, Wyschne-

gradsky’s scale can be generalized to N -tone equal temperament in
the following way. Let k be an integer k ≥ 2. The set

Wk = {0, 1, 3, 5, . . . , 2k + 1, 2k + 2, 2k + 4, . . . , 4k + 2}

is the generalized diatonic scale with |Wk | = 2k + 3 notes, and gen-

erator m = 2k + 1 in N -tone equal universe, with

N = m + |Wk | = 4k + 4

The scale Wk can be considered as a generalized major scale since

Wk contains one chord of limited transposition {0, 2k + 2}.

3.5 Exercises

Exercise 3.1
Compute the hierarchy of the well-formed scales for N = 12, m = 7
and k from 2 to 12 starting at x = 5. Compute Douthett’s J-sets
7
J12,k for k = 2, . . . , 7. Compare with the well-formed scales hierarchy.
7
Show that J12,7 is the major scale.
Exercise 3.2
Consider the C harmonic minor scale X = {0, 2, 3, 5, 7, 8, 11}, the
ascending C melodic minor scale Y = {0, 2, 3, 5, 7, 9, 11} and the
descending melodic minor scale Z = {0, 2, 3, 5, 7, 8, 10}. Are they
maximally even? Which scales have Myhill’s property?
Exercise 3.3
Consider the scales X = {0, 2, 3, 5, 6, 8, 9, 11} and Y = {0, 3, 6, 10}.
Are they maximally even?
Exercise 3.4
Consider the subset X = {0, 2, 4} of the whole tone scale
S = {0, 2, 4, 6, 8, 10}. Prove that cardinality equals variety and struc-
ture implies multiplicity are false.
Exercise 3.5
Consider the X = {0, 2, 4} of the pentatonic scale S = {0, 2, 4, 7, 9}.
Prove that cardinality equals variety and structure implies multiplic-
ity are true.
 Generalized Diatonic Scales 75

Exercise 3.6
For N = 13 and for the generator m = 8, compute the generated scale
X of k = 8 notes. Show that X is a set of progressive transposition.
8
Compute the Douthett’s scale J13,8 and connect it to X. Deduce that
X is maximally even. Compute the value of u and v using 3.1 and
3.2. Deduce that X is a diatonic scale of the 13-EDO.
Exercise 3.7
For N = 16 and for generator m = 9, compute the generated scale
X of 9 notes. Show that X is a PT set and a diatonic scale. Deduce
that X is also the generalized major scale.

References

Agmon, E. (1989). A mathematical model of the diatonic system, Journal of

Music Theory 33, 1–25.
Agmon, E. (2013). The Languages of Western Tonality, Springer, Berlin.
Balzano, G. J. (1980). The group-theoretic description of 12-fold and microtonal
pitch systems, Computer Music Journal 4, 66–84.
Broué, M. (2001). Les tonalités musicales vues par un mathématicien. In:
Rousseau, D. and Morvan, M. (dir.) Le temps des savoirs, Le code, Paris,
Odile Jacob, Revue de l’Institut universitaire de France 4, 41–82.
Carey, N. and Clampitt, D. (1989). Aspects of well-formed scales, Music Theory
Spectrum 11, 187–206.
Clough, J. and Douthett, J. (1991). Maximally even sets, Journal of Music Theory
35, 93–173.
Douthett, J. and Kranz, R. (2000). Construction and interpretation of equal-
tempered scales using frequency ratios, maximally even sets, and P-cycles,
The Journal of the Acoustical Society of America 107, 2725–2734.
Jedrzejewski, F. (2008). Generalized diatonic scales, Journal of Mathematics and
Music 2(1), 21–36.
Johnson, T.A. (2008). Foundations of Diatonic Theory: A Mathematically Based
Approach to Music Fundamentals, Key College Publishing, Emeryville
(CA).
Louvier, A. (1997). Recherche et classification des modes dans les tempéraments
égaux, Musurgia 4(3), 119–131.
This page intentionally left blank
 Chapter 4

Voice Leading and Neo-Riemannian

Transformations

In the neo-Riemannian theory, the use of algebraic structures

provides a deep insight into the concept of musical structures
and transformational processes. The contextual transformations
P (Parallel), L (Leading Tone) and R (Relative) rediscovered by
Lewin (1982), Hyer (1989, 1995) and Cohn (1997, 1998, 2012) from
the works of musicologist Hugo Riemann in the late 19th century
(Riemann, 1887) laid the foundations of the neo-Riemannian theory.
In more recent years, new theories have emerged such as the use
of orbifolds. We refer the reader to the work of Dmitri Tymoczko
(2011).

4.1 Isographic Networks

We follow in the first section the results obtained by David Lewin in

Music Theory Spectrum (Lewin, 1990). A Klumpenhouwer network
(or simply K-net) is “any network that uses T and/or I operations
to interpret interrelations among pcs.”
Definition 4.1. A Klumpenhouwer network (or simply K-net) is a
triple (V, E, G), where (V, E) is a graph and G is a finite group acting
on V . The vertices V are pitch class sets, and edges E are labeled
by elements of G. Two vertices x and y ∈ V are linked by an edge if
there exists an element g ∈ G such that y = gx. Most of the time,
G is the T /I group of transpositions and inversions and the set E is

77
78 A Compendium of Musical Mathematics

restricted and does not correspond to the set of all the actions of G
on V , for musical purposes.

Definition 4.2. Two K-nets are isographic if

(1) they must have the same configuration of vertices and edges,
(2) there must be some isomorphism F that maps the
transformation-system used to label the edges of one network
into the transformation-system used to label the edges of the
other,
(3) the transformation X labels an edge of the one network, then the
transformation F (X) labels the corresponding edge of the other.

Definition 4.3. Networks that contain identical graphs are said to

be strongly isographic.

Definition 4.4. Two K-networks are positively isographic when

(1) they share the same configuration of vertices and edges,

(2) the T -numbers of corresponding edges are equal,
(3) the I-numbers of corresponding edges differ by some fixed num-
ber j mod 12.

Definition 4.5. Two K-networks A and B are negatively iso-

graphic if

(1) they share the same configuration of vertices and edges,

(2) each T -number of B is the complement of the corresponding
T -number in A,
(3) each I-number of B is exactly j more than the I-number in A,
for some fixed number j mod 12.

4.2 Automorphisms of the T /I Group

We will see that for the T /I group there are four types of automor-
phisms which define four types of isographies. Two types of isogra-
phies are said to be positive and negative. In this section and in
the next one, we follow David Lewin (1990) where one can find the
demonstrations.
 Voice Leading and Neo-Riemannian Transformations 79

Theorem 4.1. Let u be 1, 5, 7 or 11 mod 12 and j any number mod

12. An automorphism of the T /I group is given by
F (u, j)(Tn ) = Tun F (u, j)(In ) = Iun+j
Proof. To simply the notation, F stands for F (u, j). F is an auto-
morphism if it verifies for all X, Y ∈ T /I, F (XY ) = F (X)F (Y ).
There are four generic cases:
(1) If X = Tn , Y = Tm ,
F (XY ) = F (Tn Tm ) = F (Tn+m ) = Tu(n+m)
= Tun Tum = F (X)F (Y )
(2) If X = Tn and Y = Im , then
F (XY ) = F (Tn Im ) = F (In+m ) = Iu(n+m)+j
= Tun Iun+j = F (X)F (Y )
(3) If X = In and Y = Tm , then
F (XY ) = F (In Tm ) = F (In−m ) = Iu(n−m)+j
= Iun+j Tm = F (X)F (Y )
(4) If X = In and Y = Im , then
F (XY ) = F (In Im ) = F (Tn−m ) = Tu(n−m)
= Tun+j−um−j = Iun+j Ium+j = F (X)F (Y )

Theorem 4.2. Let G be an automorphism of the T /I group. Then
there exist u = 1, 5, 7, 11 mod 12 and j such that G = F (u, j).
Proof. Since the only elements of order 12 are the translations Tu ,
with u = 1, 5, 7, 11, we have G(T1 ) = Tu , for some u ∈ {1, 5, 7, 11}.
Then G(T2 ) = G(T1 )G(T1 ) = T2u . And by induction, for any n,
G(Tn ) = Tun = F (u, j)(Tn ). For the inversions, it is sufficient to
show that for any n,
G(In ) = G(Tn )G(I0 ) = Tun Ij = Iun+j = F (u, j)(In )
which is always true if G(I0 ) = Ij for some j. Since I0 is of order 2,
G(I0 ) is also of order 2 and is either an inversion or the translation
T6 . But since G(T6 ) = T6 and G is one-to-one, G cannot map both
I0 and T6 to T6 . Hence, G(I0 ) = Ij for some j. 
80 A Compendium of Musical Mathematics

Theorem 4.3. The inner automorphisms of the T /I group are the

automorphisms F (u, j) such that u = 1 or 11 and j = 2k is even.
Proof. Since the inner automorphisms are the automorphisms
that arise from conjugation gXg−1 , we have to consider the case
g = Tk and the case g = Ik . In the first case G(X) = Tk XT−k , for
X = Tn , G(Tn ) = Tk Tn T−k = Tn and for X = In ,
G(In ) = Tk In T−k = T2k In = In+2k = F (1, 2k)(In )
In the second case G(X) = Ik XIk , for X = Tn , G(Tn ) = Ik Tn Ik =
T−n and for X = In ,
G(In ) = Ik In Ik = I2k−n = F (11, 2k)(In )
G is therefore the automorphism F (11, 2k). 

Theorem 4.4. The combination of two automorphisms F (u, j) and

F (v, k) is the automorphism F (uv, uk + j)
F (u, j)F (v, k) = F (uv, uk + j)
Proof. By computing the effect of the automorphisms on each Tn ,
F (u, j)F (v, k)(Tn ) = F (u, j)(Tvn ) = Tuvn
and on each In ,
F (u, j)F (v, k)(In ) = F (u, j)(Ivn+k ) = Iuvn+uk+j
which is exactly the definition of the automorphism F (uv, uk + j).

Applying these results to K-nets leads to the following theorem.
There are exactly four types of K-nets depending on the value of the
parameter u.
Theorem 4.5. Two K-nets are isographic if and only if there exist
an automorphism of the T /I group F (u, j) for u = 1, 5, 7 or 11 mod
12 and j ∈ Z12 . The automorphisms F (1, j) correspond to positively
isographic K-nets, and the automorphisms F (11, j) correspond to
negatively isographic K-nets.
 Voice Leading and Neo-Riemannian Transformations 81

4.3 Automorphisms of the T /M Group

Definition 4.6. The Morris T /M group is the group of the 48 pitch

class operations Tn Mu , where n = 0, 1, . . . , 11 and u = 1, 5, 7, 11
mod 12. Tn are the translations Tn (x) = x + n and Mu are the
multiplications Mu (x) = ux mod 12.
The T /I is included in the T /M group. For u = 1, M1 is the
identity and for u = 11, M11 is the inversion. So, for all n, Tn = Tn M1
and In = Tn M11 . The product of two elements is given by

(Tn Mu )(Tm Mv ) = Tn+um Muv

Theorem 4.6. Let u, j, r ∈ Z12 , such that gcd(u, 12) = 1, j is

even and r is divisible by 3. An automorphism of the T /M group
F = F (u, j, r) is given by

F (Tn ) = Tun , F (Tn M5 ) = Tun+j M5

F (Tn M7 ) = Tun+r M7 , F (Tn M11 ) = Tun+j+r M11

Proof. We have to verify F (XY ) = F (X)F (Y ) for each of the 16

cases X = Tn Mu and Y = Tm Mv . For instance, for X = Tn M11 and
Y = Tm M11 , one has

F (XY ) = F (Tn M11 Tm M11 ) = F (Tn−m ) = Tu(n−m)

On the other hand,

F (X)F (Y ) = F (Tn M11 )F (Tm M11 ) = Tun+j+r M11 Tum+j+r M11

= Tun+j+r T−um−j−r = T u(n − m)

The demonstrations of the other cases are analogous. 

Lemma 4.1. Let X be an element of the T /M group:

(1) X has order 12 if and only if X = Tu , with u ∈ {1, 5, 7, 11}.

(2) X has order 6 if and only if X = T2 , T10 or X = Tn M7 and n
is not divisible by 3.
82 A Compendium of Musical Mathematics

(3) X has order 4 if and only if X = T3 , T9 or X = Tn M5 and n is

odd.
(4) X has order 3 if and only if X = T4 or T8 .
Proof. The verification is straightforward for all powers of trans-
lations Tn . The remaining cases Tn M5 and Tn M7 are the subject of
the following lemma. 

Lemma 4.2.
(1) Tn M5 has order 2 if n is even and order 4 if n is odd.
(2) Tn M7 has order 2 if n is a multiple of 3 and of order 6 otherwise.
Proof. From (Tn M5 )(Tn M5 ) = Tn T5n = T6n , we see that if n is
even, T6n = T0 and so Tn M5 has order 2. If n is odd, T6n = T6 and
so Tn M5 has order 4. From (Tn M7 )(Tn M7 ) = Tn+7n M1 = T8n , we
see that if n is a multiple of 3, T8n = T0 and so Tn M7 has order 2.
Otherwise, T8n = T8 if n = 1 mod 3, and T8n = T4 if n = 2 mod 3.
So, Tn M7 is of order 7. 

Theorem 4.7. Let G be an automorphism of the T /M group. Then

there exist u = 1, 5, 7, 11 mod 12, j even and r divisible by 3 such
that G = F (u, j, r).
Proof. Since G(T1 ) is of order 12, there exist u such that
G(T1 ) = Tu . As in the case of the demonstration of the same result
for the T /I group, we infer by induction that for all n, G(Tn ) = Tun .
Since Tn M7 has order 6 if n is not a multiple of 3, G(Tn M7 ) has
also order 6. And from the previous lemma, G(Tn M7 ) = Tp M7 for
some p not divisible by 3. By induction, we can show that G(Tn M7 ) =
Tun+p−u . So, there exists r = p − u such that G(Tn M7 ) = Tun+r . The
number r is divisible by 3, since

G(T0 ) = G(M7 M7 ) = Tr M7 Tr M7 = T8r

Hence, 8r = 0 mod 12 and r is divisible by 3.

Since Tn M5 has order 4, G(Tn M5 ) has also order 4. Furthermore,
since G maps transpositions into transpositions, G(Tn M5 ) cannot be
a transposition. By the lemma, G(Tn M5 ) = Tq M5 , for some odd q.
Set j = q − u, j is even and by induction, for all n,

G(Tn M5 ) = G(Tn )G(M5 ) = Tun Tj M5 = Tun+j M5

 Voice Leading and Neo-Riemannian Transformations 83

Furthermore, one has

G(Tn M11 ) = G(Tn M5 M7 ) = G(Tn M5 )G(M7 )
= (Tun+j M5 )(Tr M7 ) = Tun+j T5r M5 M7
Since r is divisible by 3, 5p = p mod 12, thus
G(Tn M11 ) = Tun+j Tr M5 M7 = Tun+j+r M11
In conclusion, the automorphism G = F (u, j, r). 

Theorem 4.8. The inner automorphisms of the T /M group are the

automorphism F (u, j, r) such that u = 1, 5, 7, 11 mod 12, j is divisible
by 4 and r is divisible by 6.
Proof. The inner automorphisms of the T /M group are the auto-
morphism F (u, j, r) such that
F (X) = (Tm Mv )X(Tm Mv )−1 = (Tm Mv )X(T−vm Mv )
From Lemma 4.2, the value of j is determined by F (M5 ) = Tj M5 ,
and
F (M5 ) = (Tm M5 )M5 (T−vm M5 ) = Tm M5v T−vm Mv
= Tm T−5m M5v Mv = T8m M5
Hence, j = 8m, so j is divisible by 4.
From Lemma 4.2, the value of r is determined by F (M7 ) = Tr M7 ,
and
F (M7 ) = (Tm Mv )M7 (T−vm Mv )
= Tm M7v T−vm Mv = T6m M7
Hence, r = 6m, where r is divisible by 6.
From Lemma 4.2, the value of u is determined by F (T1 ) = Tu ,
and
F (T1 ) = (Tm Mv )T1 (T−vm Mv ) = Tm Tv Mv T−vm Mv = Tv
Hence, u = v. We have just shown that given any inner auto-
morphisms F , F is of the type obtained for the values of (u, j, r).
Conversely, it is necessary to check that the automorphisms F found
for the values (u, j, r) correspond to all inner automorphisms, in other
84 A Compendium of Musical Mathematics

words, given u = 1, 5, 7 or 11 and j divisible by 4 and r divisible by 6,

there exists a value k ∈ Z12 such that j = 8k and r = 6k. By exam-
ining the 6 possible cases j ∈ {0, 4, 8} and r ∈ {0, 6}, it is easy to see
that we can always find two values for k satisfying the two preceding
equations. For instance, if j = 4, r = 6, k = 5 or 11. This ends the
demonstration. 

4.4 PLR Transformations

In the following sections, we follow by updating the articles

(Jedrzejewski, 2005, 2019a, 2009b). A major triad is a set of pitches
consisting of three notes that have a root x, major third x + 4
and perfect fifth x + 7. A minor triad is a set of pitches con-
sisting of three notes that have a root x, minor third x + 3 and
perfect fifth x + 7, where the root of the chord ranges through
Z12 . Major and minor triads are identified with 3-element subsets
of Z12 of the form {x, x + 4, x + 7} and {x, x + 3, x + 7}, respec-
tively. The set of the 24 major and minor chords is also called the
set of consonant triads. Their interval structure can be expressed in
terms of third chains (4, 3) and (3, 4), respectively. Rooted interval
chains ((a, b), x) can be mapped to the associated pitch class sets via
((a, b), x) → {x, x + a, x + a + b}. Some neo-Riemannian approaches
use pitch-class segments denoted by angular brackets.
Definition 4.7. The set Σ of consonant triads then consists of the
major segments x, x + 4, x + 7 and minor segments x, x − 4, x − 7
where the (dualistic) root x ranges over Z12 .
The properties on PLR groups have been established by Thomas
Fiore et al. (Berry and Fiore, 2018; Fiore and Satyendra, 2005; Fiore
and Noll, 2011, 2018; Fiore et al., 2013).
Definition 4.8. The PLR transformations on (dualistic) root posi-
tion triads within Σ are defined by
P x, y, z = Ix+z x, y, z (4.1)
R x, y, z = Ix+y x, y, z
L x, y, z = Iy+z x, y, z
 Voice Leading and Neo-Riemannian Transformations 85

Denote xM ∈ Σ the major chord rooted on x and xm ∈ Σ the

minor chord with root x, the P , L, R transformations act as follows.
The transformation P (Parallel operation) exchanges a major
triad with the associated parallel minor triad
P (xM ) = xm , P (xm ) = xM
For instance,
P c, e, g = P 0, 4, 7 = 7, 3, 0 = g, e, c
The transformation R (Relative operation) exchanges a major triad
with its relative minor triad (the real root of the relative minor triad
is a minor third below the root of the major triad)
R(xM ) = (x + 9)m
For instance,
R c, e, g = R 0, 4, 7 = 4, 0, 9 = e, c, a
And the transformation L (Leittonwechsel, Leading tone operation)
exchanges a major triad with a minor triad with its real root a major
third above
L(xM ) = (x + 4)m
For instance,
L c, e, g = L 0, 4, 7 = 11, 7, 4 = b, g, e
The transformation R is also linked to P and L as
P = (RL)3 R
In other words, P relates triads that share a common fifth, L relates
triads that share a common minor third and R relates triads that
share a common major third.
Remark 4.1. Observe, however, that applying formulae (4.1) to
chord inversions leads to L 0, 4, 7 = 11, 7, 4 (Em), L 7, 4, 0 =
9, 0, 4 (Am) and L 4, 7, 0 = 3, 0, 7 (Cm). In this case, one can-
not use the equivalence of chord inversions. But these formulae can
alternatively be interpreted as voicing transformations (Fiore and
Noll, 2018).
86 A Compendium of Musical Mathematics

Proposition 4.1. The PLR transformations are involutions P 2 =

L2 = R2 = Id; they are their own inverses.

Definition 4.9. The group generated by P , L and R is called

the P LR-group, the Schritt/Wechsel group or the neo-Riemannian
group after the late 19th-century music theorist Hugo Riemann. Since
P = R(LR)3 , the P LR group is generated by L and R.

Proposition 4.2. The PLR group is the dihedral group of 24

elements.

Corollary 4.1. The PLR group acts simply transitively on the set
of consonant triads.

In the following, major and minor chords are indicated by compact

pitch class notations with A = 10 and B = 11. Symbol 904 is the
minor chord 9, 0, 4 = e, c, a and 48B is the major chord 4, 8, 11 =
e, g, b. The planar representation of the PLR group is a torus (see
Fig. 4.2).
As remarked above, the transformations P, L, R are not defined
on the 12 pitch classes themselves. P sends e = 4 in 0, 4, 7 to 3 and
in 9, 1, 4 to 9. The transformations are contextually defined on the
basis of the specific interval structure of the major and minor triads.
There is no canonic way to define them in strict analogy on other
single chord classes, and there are serious obstacles to extend them
to all pitch class sets at once. Therefore, the goal of the following
section is to find noncontextual inversions as partners for P , L and
R on Z12 , which are then automatically valid for all k -chords.

4.5 JQZ Transformations

In this section, we use the term JQZ group synonymously to the term
T /I group. This is in analogy to the usage of the term PLR group
synonymously to the term S/W group (Schritt/Wechsel group). This
particular system of generators J, Q and Z has very similar properties
to the generators P , L and R of the Schritt–Wechsel group. They
are not contextually defined and can be applied pointwise. Their
definition is unique up to conjugation. Our concrete choice depends
upon the choice of the C-major triad {0, 4, 7} as a distinctive chord of
reference. The choice of any other consonant triad f ({0, 4, 7}) in this
 Voice Leading and Neo-Riemannian Transformations 87

role yields a conjugated triple f Jf −1 , f Qf −1 , f Zf −1 of generators.

The J, Q, Z transformations have been introduced in Jedrzejewski
(2005).
The purpose of this section is to select congenial inversions for
the generators P , L and R among the 12 inversions Ik on Z12 .
In the first step, we look for an inversion J that behaves like the
transformation P for the C-major triad, i.e. which transforms the
major chord {0, 4, 7} into the minor chord {0, 3, 7}. The transforma-
tion J = I7 fulfills this and is uniquely determined. For all x ∈ Z12 ,
we get J({x, x + 4, x + 7}) = {−x, −x + 3, −x + 7}. In other words,
J transforms the major chord rooted at x into the minor chord rooted
at −x (mod 12). It can be interpreted as the permutation (in cyclic
notation): J = (0, 7)(1, 6)(2, 5)(3, 4)(8, 11)(9, 10).
In the second step, we look for an inversion Q that behaves
like the Leitonwechsel L on the C-major triad, i.e. transform-
ing the major chord {0, 4, 7} into the Leading-tone-exchange chord
(Leittonwechselklang) {4, 7, 11}. The transformation Q = I11 ful-
fills this and is uniquely determined. For all x ∈ Z12 , we get
Q({x, x + 4, x + 7}) = {−x + 4, −x + 7, −x + 11}. In other words,
Q transforms the major chord rooted at x into the minor chord
rooted at 4 − x (mod 12). The transformation Q is the permuta-
tion Q = (0, 11)(1, 10)(2, 9)(3, 8)(4, 7)(5, 6).
In the third and last step, we look for an inversion Z which
behaves like the transformation R on the C-major triad, i.e. which
transforms the major chord {0, 4, 7} into the relative minor chord
{9, 0, 4}. The transformation Z = I4 fulfills this and is uniquely deter-
mined. For all x ∈ Z12 , we get Z({x, x+4, x+7}) = {−x+9, −x, −x+
4}. In other words, Z transforms the major chord rooted at x into
the minor chord rooted at 9 − x (mod 12). The transformation Z is
the permutation: Z = (0, 4)(1, 3)(5, 11)(6, 10)(7, 9).

Definition 4.10. The JQZ transformations are defined on Zk12 by

J = I7 , Q = I11 and Z = I4 .

Since JQZ transformations act pointwise, the order of the pitch

classes does not matter.
As noted, the transformations J, Q, Z depend upon the choice
of the C-major triad. But analogous transformations can be chosen
with a different chord of reference, such that they behave like the
88 A Compendium of Musical Mathematics

transformations P, L, R on this chord. Introducing the transforma-

tions J(x, +) = I2x+7 , Q(x, +) = I2x−1 , Z(x, +) = I2x+4 for a major
chord and the transformations J(x, −) = I2x+7 , Q(x, −) = I2x+3 ,
Z(x, −) = I2x−2 for a major chord, we have the following result.
Theorem 4.9. For the major chord X = {x, x + 4, x + 7}, we define
the inversions
J(x, +) = I2x+7 , Q(x, +) = I2x−1 , Z(x, +) = I2x+4

For the minor chord X = {x, x + 3, x + 7}, we define the inversions

J(x, −) = I2x+7 , Q(x, −) = I2x+3 , R = Z(x, −) = I2x−2

In both cases s = ±, we obtain J(x, s)(X) = P (X), Q(x, s)(X) =
L(X), Z(x, s)(X) = R(X). Moreover, the transformations J(x, s),
Q(x, s), Z(x, s) are conjugates of J, Q and Z, respectively:

J(x, +) = Tx JTx−1 , Q(x, +) = Tx QTx−1 , Z(x, +) = Tx ZTx−1 ,

J(x, −) = Ix JIx , Q(x, −) = Ix QIx , Z(x, −) = Ix ZIx

Proof. The proof is straightforward using Equations (4.1). For

instance, one has

P x, x + 4, x + 7 = I2x+7 x, x + 4, x + 7 = J(x, +)

Using the properties (see Equations (4.2)), we have

Tx JTx−1 = Tx I7 T−x = Tx I7+x = I2x+7 = J(x, +) = J(x, −)

Other relationships are shown in the same way. 
These conjugation relations allow us to choose a reference point
(here x = 0), but it can be adapted for musical purpose. The Cayley
graph of the action of the neo-Riemannian group on the consonant
triads with respect to the generators P , L and R is known under
the nickname “chicken-wire torus”. Analogously, the Cayley graph
of the JQZ transformations is represented in Fig. 4.1. The torus is
visualized by gluing the right border with the left side and the top
side with the bottom, according to the chords.
Proposition 4.3. The JQZ transformations are involutions J 2 =
Q2 = Z 2 = Id.
 Voice Leading and Neo-Riemannian Transformations 89

Fig. 4.1. Tonnetz of the JQZ -transformations.

Definition 4.11. The JQZ group is the group of order 24, generated
by the JQZ transformations, with presentation:

J, Q, Z = J, Q, Z | J 2 = Q2 = Z 2 = (JQ)3

= (JZQ)2 = (JZ)4 = 1

The toroidal representation was made by several authors including

Richard Cohn (1997, 1998, 2012). Derek Waller was one of the first
authors to introduce the torus in musical representations (Waller,
1978). The donut that is made by gluing the edges of the tonnetz
was called the chicken-wire torus by Douthett and Steinbach (1998).
Whether we start from the P LR network or from the JQZ network,
we find the same topological figure (see Fig. 4.2).
In the Cube Dance, another figure introduced by Douthett and
Steinbach (1998), each vertex represents a consonant triad or an
augmented triad, and each solid edge is labeled by either P or L.
A slightly differing figure can be obtained with the transformations
J and Q (see Fig. 4.3).
90 A Compendium of Musical Mathematics

Fig. 4.2. Chicken-wire torus (left P LR, right JQZ).

Fig. 4.3. Douthett’s cube dance (left P LR, right JQZ).

Although there is an isomorphism between PLR and JQZ groups,

this isomorphism is not obvious. In the T/I group, the transpositions
commute Tn Tm = Tm Tn = Tm+n but the inversions do not commute.
They satisfy the relations
In Im = Tn−m Tn Im = In+m Im Tn = Im−n (4.2)
Transformations P, L, R commute with transpositions (P Tn = Tn P,
LTn = Tn L, RTn = Tn R), while J, Q and Z anticommute (JTn =
T−n J, QTn = T−n Q, ZTn = T−n Z).
Theorem 4.10. The transformations J, Q, Z satisfy the relations
JIn = I2−n J, ZIn = I8−n Z, QIn = I10−n Q
Commutation relations are obtained for JI1 = I1 J, ZI10 = I10 Z,
QI5 = I5 Q.
 Voice Leading and Neo-Riemannian Transformations 91

Proof. Substituting J, Q, Z by respectively I7 , I11 , I4 leads to the

relations. For instance,
JIn = I7 In = T7−n = T2−n+5 = T2−n−7 = I2−n I7 = I2−n J
Commutation relations are obtained only if Im In = T6 . Since Im In =
Tm−n , n must be equal to n = m + 6 mod 12. Then, for m = 7, n = 1,
for m = 10, n = 4 and for m = 4, n = 10. 
Theorem 4.11. The transformations P , L, R on the set of conso-
nant triads Σ satisfy the relations
Tx−y P L = Ty−x LP, Tz−y P R = Ty−z RP, Tz−x LR = Tx−z RL
Proof. We have
P L x, y, z = P y + z − x, z, y = y, 2y − x, y + z − x
On the other hand, the computation
T2y−2x LP x, y, z = T2y−2x L z, x + z − y, x
= T2y−2x 2x − y, x, x + z − y
= y, 2y − x, y + z − x = P L x, y, z
leads to the formula. The proof of the two other relations is similar.

From the musical point of view, the JQZ and PLR paths in the
tonnetz are very similar. The examples studied by Crans et al. (2009)
are reinterpreted here by JQZ relations. For instance, in the “Grail”
theme of the Prelude of Parsifal (1882), the chord progression writ-
ten under the compact form
A Fm D Bm A
→ → → →
803 580 158 A15 803
is interpreted in two ways. The PLR interpretation shows the impor-
tance of relative chords while the JQZ interpretation highlights the
action of the octatonic subgroup, namely the inversion ZJZ between
chords. The vertical arrows are not the same in the PLR transfor-
mations (one is R and the other RLR) while they are perfectly sym-
metrical (ZJZ = J Z ) in the case of the group JQZ, representing the
inversion I1 . Taking the chord A as a reference will lead to simpler
relationships.
92 A Compendium of Musical Mathematics

PLR Interpretation: JQZ Interpretation:

R QJZ
A Fm A Fm
RLR L ZJZ ZJZ

Bm D Bm Db
R QZQ

In the Lento occulto of Feruccio Busoni’s Sonatina seconda, the

similarity between PLR and JQZ groups is clear in the chord
progression:

ZJZQ
E / C ZJQJ / D ZJQJ/ E ZJQJ / E
O RP RL P RP L P RP L P RP L
RL QZ
JQJZ QJ QZ 
E o Eo Co F o JZJZ B
P LP R PL RL P RP R

as well as in Debussy’s Danseuses de Delphes (Preludes vol. 1):

ZQ
B
J
Dm / Gm JQZ C ZQZ Dm J
B
ZQZQ
/ C ZQ / F
L RL P RL RLR L RLRL LR

4.6 Neo-Riemannian Groups

As in the case of the P LR group, two subgroups of the JQZ group are
important: the hexatonic group generated by transformations J and
Q and the octatonic group generated by transformations J and Z.

Definition 4.12. The hexatonic group is the group generated by

transformations J and Q. It has the presentation
 
J, Q = J, Q | J 2 = Q2 = 1, JQJ = QJQ

The group has six elements: J, Q = {T0 , I7 , T4 , I3 , T8 , I11 } =

{1, J, QJ, JQJ, JQ, Q}.

Each element of the group is computed using the relations J = I7

and Q = I11 . For instance,

JQ(x) = J(−x+11) = −(−x+11)+7 = x−4 = x+8 mod 12 = T8 (x)

 Voice Leading and Neo-Riemannian Transformations 93

The group acts on the major and minor chords, but likewise it acts
on any other set class. The 3-chord {0, 6, 11} has the 6-element orbit

{{0, 6, 11}, {1, 7, 8}, {3, 4, 10}, {3, 4, 9}, {2, 7, 8}, {0, 5, 11}}

within the set class 3–5 (Forte’s nomenclature).

Definition 4.13. The octatonic group is generated by the transfor-
mations J and Z. It has the presentation
 
J, Z = J, Z | J 2 = Z 2 = 1, (ZJ)2 = (JZ)2

The group has eight elements:

J, Z = {T0 , I7 , T9 , I10 , T6 , I1 , T3 , I4 }

= {1, J, ZJ, JZJ, (ZJ)2 , ZJZ, JZ, Z}

For instance, we compute

JZ(x) = J(−x + 4) = −(−x + 4) + 7 = x + 3 mod 12 = T3 (x)

Since the groups T/I, PLR and JQZ are isomorphic, they have
the same center. This centre is the group of order 2 consisting of
{1, T6 }.
Introduced by Hook (2002), the group of uniform triadic trans-
formations (U T T ) acts on the set of 24 major and minor triads Σ.
Its elements are denoted by (p, q, σ) with p, q ∈ {0, 1, . . . , 11} and
σ = ±1. Let (x)M be a major triad rooted on x and (x)m a minor
triad rooted on x, the (p, q, +1) U T T elements send the major triad
(x)M to (x+p)M and the minor triad (x)m to (x+q)m . The (p, q, −1)
U T T elements send the major triad (x)M to the minor triad (x+p)m
and the minor triad (x)m to the major triad (x + q)M .
Proposition 4.4. The group U of uniform triadic transformation
(UTT) of order 288 is isomorphic to the wreath product of the cyclic
group C12 of order 12 by the cyclic group C2 of order 2:

U = C12  C2

The group Q of quasi uniform triadic transformations of order

1152 is a refinement of the group U of uniform triadic transformations
of order 288. This can be easily checked using the GAP software.
94 A Compendium of Musical Mathematics

Definition 4.14. The group Q of quasi uniform triadic transfor-

mation (QUTT) is the wreath product of the T/I group by the
symmetric group of two elements S2 :
Q = T /I  S2  D12  S2
The notion of dual groups in the sense of Lewin has been intro-
duced in Fiore and Noll (2011).
Definition 4.15. Two subgroups G1 , G2 of a group S are dual (in
the sense of Lewin) if both act simply and transitively and are each
others centralizers:
CS (G1 ) = G2 and CS (G2 ) = G1
For instance, the centralizer of the T/I group in Q is the JQZ
group (or PLR group) and the centralizer of the JQZ group (or PLR
group) in Q is the T/I group. The computation is straightforward
with the GAP software:
CQ (T /I) = J, Q, Z and CQ (J, Q, Z) = T /I
The same is true in the symmetric group Sym(Σ)  S24 of permu-
tations of 24 elements (Crans et al., 2009) instead of Q: the JQZ
group and the T/I group are subgroups of Sym(Σ). The subduality
of the hexatonic group and the octatonic group has been studied by
Thomas Fiore and Thomas Noll in the same reference (Fiore and
Noll, 2011). They define the concept of subduality in the following
sense.
Definition 4.16. Let G1 , G2 be dual subgroups of the symmetric
group S24 of the 24 triads and let H1 , H2 be two subgroups of G1 ,
G2 , respectively. H1 and H2 are subdual groups if both act simply
and transitively on a subset X ⊂ S24 and are each others centralizers
within the symmetric group SX of this subset:
CSX (H1 ) = H2 and CSX (H2 ) = H1
The hexatonic JQ group J, Q has 4 orbits, namely two pairs of
tritone-related sets of six triads each:
{C, Cm, E, Em, A, Am} and {D, Dm, F , F m, B , B m} as
well as {D , E m, F, Gm, A, Bm} and {D m, E , Fm, G, Am, B}.
 Voice Leading and Neo-Riemannian Transformations 95

The orbits in the first pair are triads in proper hexatonic collections,
i.e. they are also orbits of the hexatonic PL-group P, L. The orbits
in the second pair form orbits under a conjugate of P, L, namely
R P, L R. This leads to the following result.
Theorem 4.12. With respect to the first two orbits, the group J, Q
is subdual to P, L. With respect to the second two orbits, J, Q is
subdual to R P, L R.
The octatonic JZ group J, Z has three orbits of eight triads
each. The first one {F m, F , E m, A, Cm, C, Am, E } is also the
orbit of the octatonic P R group. The two others {Bm, C , Gm, E,
Fm, G, Dm, B} and {Bm, D, Gm, F, Em, G, C m, B } are orbits
under conjugation (RL) P, R (RL)−1 . This lead to the following
result.
Theorem 4.13. With respect to the first orbit, the group J, Z
is subdual to P, R. With respect to the second two orbits, J, Z is
subdual to (RL) P, R (RL)−1 .
Some other groups have been introduced to generalized neo-
Riemannian transformations. Alexander Popoff (2013) defined the
Z3  Q8 groups as follows, where Q8 is the quaternion group of
presentation:
 
Q8 = i, j, k | i2 = j 2 = k 2 = ijk
It has also a presentation with two generators:
 
Q8 = u, v | u4 = 1, u2 = v 2 , uvu = v
This group acts simply transitively on the set of major and minor
triads. It is generated by the transformations u and v such that
u(xM ) = (x + 1)M u(xm ) = (x − 1)m
v(xM ) = xm v(xm ) = (x + 6)M
Verification of relationships is obvious. For instance, for the last
one, we have
uvu(xM ) = uv(x + 1)M = u(x + 1)m = xm = v(xM )
uvu(xm ) = uv(x − 1)m = u(x + 5)M = (x + 6)M = v(xm )
96 A Compendium of Musical Mathematics

4.7 Atonal Triads

The study of the interplay of the PLR group and the JQZ group is
helpful in atonal analysis. We will illustrate this for the the atonal
triad 0, 1, 6.
The main advantage of JQZ transformations is to be able to con-
sider all types of chords and not only consonant chords. For instance,
if the JQZ group acts on the atonal triad 016 {c, c, f }, the action
leads to a new lattice. In the other hand, if we replace the set Σ of
consonant triads by the set Γ of atonal triads rooted on x of the form
x, x + 1, x + 6 and x, x − 1, x − 6, we get new relations

P x, x + 1, x + 6 = x + 6, x + 5, x ,
P x, x + 11, x + 6 = x + 6, x + 7, x
R x, x + 1, x + 6 = x + 1, x, x + 7 ,
R x, x + 1, x + 6 = x + 11, x, x + 5
L x, x + 1, x + 6 = x + 7, x + 6, x + 1 ,
L x, x + 11, x + 6 = x + 5, x + 6, x + 11 .

This allows us to interpret George Crumb’s Gargoyles, a piano piece,

extract from Makrokosmos, in both ways:
2
238
JSZ / 56B QZ / 016 JZ / 349 QJ / 781 (JZ) / 127
(P L)3 PR (P L)4 LP RP RL

2
107
JZ / 43A QZ / BA5 JZ / 218 QJ / 650 (ZJ) / 0B6
(LP ) 3 RP 3(LP ) 4 (LP ) LR

The right hand evolves in the same way as the left hand: the
atonal triads are linked by the transformations (JZ, QJ). The same
transformations are used in both hands, except in the last triads.
But if we consider a crossing of the hands, 781 goes to 0B6 by the
transformation J, just as 650 goes to 127 by the same transformation.
Another interpretation is possible. As we saw for the PLR group,
we can consider two kinds of atonal triads: one of structure (1, 5) and
the other of structure (5, 1). From the relations of P LR seen above,
we can introduce the transformations for atonal triads {x, x+1, x+6},

J(x, +) = I2x+6 , Q(x, +) = I2x+7 , Z(x, +) = I2x+1

 Voice Leading and Neo-Riemannian Transformations 97

and for the atonal triads {x, x + 5, x + 6},

J(x, −) = I2x+6 , Q(x, −) = I2x−7 , Z(x, −) = I2x−1
For x = 0, we have three new transformations J = I6 , Q = I7 ,
Z = I1 . The tritone (6) has replaced the fifth (7) in the definition of
J. Applying these transformations to Olivier Messiaen’s Regard de
l’Onction terrible (extract of Vingt Regards de l’Enfant Jesus #18)
leads to the following chords progression, which evolves against the
same chord A94, on the left hand. We use the following notation for
the conjugation: Z J = JZJ.

JQ JQ JQ JQ JQ
650 / 54B / 43A / 329 / 218 / 107 / 107
PL PL PL PL PL
4
RP L ZJ Q JZ RP ZQ RL ZJ RP L (LP ) ZQJ (LP )3 ZJ J Q J Q (RP )2
      JQ 
A94 / A94 / A94 / A94 / A94 / A94 / 0B6
LP

This demonstrates the interest and power of these noncontextual

transformations for the analysis of all kind of music, especially for
atonal music.

4.8 Seventh Chords

In addition to triads, seventh chords are most prominent in the har-

monic vocabulary of tonal music. Initially, neo-Riemannian theory
was concerned only with the set of consonant triads. Later, some
authors have introduced new transformations to consider seventh
chords and have developed new generalizations of the well-known
P LR transformations. For instance, we propose a set of transfor-
mations based on third chains and we introduce new groups acting
on seventh chords, ninth chords or more generally on any chain of
major and minor thirds. Considering stacks of thirds as Jean-Philippe
Rameau did in 1722 to define chords, and taking into account that
these stacks of thirds are defined in Z12 , we introduce the Rameau–
Schillinger operators and study the hierarchy of groups that they
generate.
As we have just noticed, several authors proposed to extend the
P LR transformations to the case of seventh chords. At the end of
the 20th century, Edward Gollin (1998) and Adrian P. Childs (1998)
studied some relations between seventh chords in three-dimensional
98 A Compendium of Musical Mathematics

expansions of the tonnetz. They incorporated the dominant sev-

enths and the half-diminished seventh chords into the scope of neo-
Riemannian theory. Childs considered examples from Chopin and
Wagner. More recently, the case of seventh chords was also studied
by Arnett and Barth (2011), Boris Kerkez (2012) as well as Sonia
Cannas (Cannas et al., 2017; Cannas, 2018). In 2012, Boris Kerkez
(2012) proposed to extend PLR transformations to major and minor
seventh chords. He defines two maps from the set H of major and
minor seventh chords to itself:

P (x, y, z, t) = (z, y, x, 2σ(t) + t)

S(x, y, z, t) = (x − 2σ(t), t, z, y)

where σ(t) = −1 if t is a major seventh and σ(t) = +1 if t is a minor

seventh. The chords are supposed to be oriented and all computations
are done modulo 12. Kerkez showed that the group generated by P
and S is isomorphic to the dihedral group D24 of order 24.
Cannas found a group of operations between seventh chords capa-
ble to describe parsimonious voice leading, moving a single note by
a semi-tone or a whole tone. She considered the P LRQ group gen-
erated by 17 operators (Cannas et al., 2017; Cannas, 2018) acting
on five types of seventh chords. She showed that the P LRQ group
is isomorphic to the semi-product of the symmetric group of permu-
tation S5 by Z412 . In addition, she proposed a new scheme for the
power towers graph, introduced by Douthett and Steinbach (1998).
Moreover, she generalized her theory to 9 types of seventh chords
and defined a new group P LRQ∗ generated by 37 operators. This
group P LRQ∗ is isomorphic to S9  Z812 . But these two groups, the
P LRQ and the P LRQ∗ groups, are very large.
In the following, we consider seven types of seventh chords (see
Table 4.1). These chords are stacks of major (4) or minor thirds (3).
The chords are marked by their intervallic structure. For instance,
433 denotes dominant seventh chords.
Transpositions Tn retain the intervallic structure of seventh
chords. But the inversions In act differently depending on the type
of seventh chords. Under the action of In , a dominant seventh chord
(433) is transformed into a half-diminished seventh chord (334).
A minor major seventh chord (344) is transformed into an augmented
major seventh chord (443). The three other types of seventh chords
 Voice Leading and Neo-Riemannian Transformations 99

Table 4.1. List of the seven types of seventh chords.

Dominant Seventh 433 C7 047A

Half-Diminished Seventh 334 Cm75 036A
Major Seventh 434 CM7 or CΔ 047B
Minor Seventh 343 Cm7 037A
Diminished Seventh 333 Cdim7 0369
Minor Major Seventh 344 CmM 7 037B
Augmented Major Seventh 443 Cmaj75 048B

retain their intervallic structure under the action of inversion. The

set of diminished seventh chords forms a limited transposition set.
There exist only three diminished seventh chords: 036A, 147A and
258B. They are their own inverses.
In the sequel, we will interpret relations between the seven types
of seventh chords in terms of transformations, which shall be called
Rameau–Schillinger operators (Jedrzejewski, 2009b):
• R1 exchange the first letter of the structure: 4 → 3 and 3 → 4 (R1
is an involution R12 = 1): 
334 → 434, 
433 → 333.
• R2 exchange the second letter of the structure: 4 → 3 and 3 → 4
(R2 is an involution R22 = 1): 3
33 → 343.
• R3 exchange the third letter of the structure: 4 → 3 and 3 → 4
(R3 is an involution R32 = 1): 34
3 → 34 4.
The three Rameau–Schillinger operators verify the relations
R12 = R22 = R32 = 1
(R1 R2 )2 = (R2 R3 )2 = (R1 R3 )2 = 1
In the following, we will use to simplify the notations Rij = Ri Rj ,
Rijk = Ri Rj Rk , etc. Let us observe for the following that the only
Rameau–Schillinger operator that is an inversion is the operator R13 .
The group generated by R1 , R2 and R3 is commonly called the ele-
mentary Abelian group of order 8. This group has the following pre-
sentation which uses precisely the relations found above:
 
R1 , R2 , R3 | R12 = R22 = R32 = (R1 R2 )2
E8 =
= (R2 R3 )2 = (R1 R3 )2 = 1
It acts on the set of the seven types of seventh chords through
the transformation of their intervallic structures. Adding chord
100 A Compendium of Musical Mathematics

fundamentals and the semitone transposition T = T1 , the group

generated by R1 , R2 , R3 , T acts on the set of all seventh chords.
The relationships of seventh chord types through Rameau–Schillinger
operators are shown in Fig. 4.4. This figure shows how the action of
Rameau–Schillinger operators on the seventh chord types is exem-
plified by those chords which are rooted on the pitch class 0.
As the augmented triad with three major thirds appears as a limit
point (see Fig. 4.4), we need to add this chord and its transpositions
to the seventh chord set. We therefore consider the set U containing
all seven types of seventh chords as well as the four augmented triads.
The set U represents 8 types of intervallic structures and has 79
elements (since there are only 3 diminished seventh chords and 4
augmented triads). Elements X of U are of the form X = {x, x +
a, x + a + b, x + a + b + c} with a, b, c ∈ {3, 4}. But in order to apply
the operators {Rj }j=1,2,3 consistently, we consider also the set U  of
rooted third chains
 = {3, 4}3 × Z12
U
of 96 elements. This set represents seventh chords as third chains
(a, b, c) ∈ {3, 4}3 which are rooted in some pitch class x ∈ Z12 . We
recover the set U by the projection

π : U−→ U, π((a, b, c), x) = {x, x + a, x + a + b, x + a + b + c}

Fig. 4.4. Action of the Rameau–Schillinger operators on the seventh chords.

 Voice Leading and Neo-Riemannian Transformations 101

Consider now a simple example. In the seventh chords progression

described as follows:

T2 R123 T2 T1 R123
EM 7 / F m7 / Gm7 / AM 7 T11 R123/ Gm7 T10
/ F m7 T10 R123/ EM 7

the axis of symmetry of the palindromic progression passing through

the chord AM 7 shares an ascending move from EM 7 to AM 7 and
a descending move from AM 7 to EM 7 . The moves from major
seventh chord to minor seventh chord or from minor seventh chord
to major seventh chord are represented by the operator R123 which
among the Rameau–Schillinger operators is the only operator used
in this excerpt (apart from the identity operator).
Rameau–Schillinger operators act on the set U.  To see how the
group acts, it is enough to transpose Fig. 4.4 on each note of the 12
tone scale.
Definition 4.17. The seventh chord group E8 × C12 is the direct
product
 of the elementary
 Abelian group E8 by the cyclic group
C12 = T | T 12 = 1 . The group has the following presentation:

RC3 = E8 × C12

 
R1 , R2 , R3 , T | R12 = R22 = R32 = T 12 = (R1 R2 )2
= = (R2 R3 )2 = (R1 R3 )2 = T −1 R1 T R1
= T −1 R2 T R2 = T −1 R3 T R3

Theorem 4.14. The seventh chords group E8 × C12 is an Abelian

group of order 96, isomorphic to C2 × C2 × C2 × C12 . The group acts

simply and transitively on U.
Proof. As the product of two Abelian groups, the group E8 × C12
is Abelian. The group has order

|RC3 | = |E8 | × |C12 | = 8 × 12 = 96

The group E8 is isomorphic to C2 × C2 × C2 which implies that the

group RC3 is isomorphic to C23 × C12 . 
In Waltz for Debby of Bill Evans, the music starts on a major
seventh chord (434) and the transpositions are at the minor third
down (T9 ) or a fourth up (T5 ). The analysis distinguishes the changes
102 A Compendium of Musical Mathematics

of chord types from simple transpositions or inversions. Only one

sequence from Gm75 to C7 (T5 R13 ) is an inversion (I5 ):
T9 R123 T5 T9 R12 T5 R3 T5 R3
F Δ /A / Dm7 / Gm7 / E 7 /G / A7 /G / D7 /F 

T5 T5 T5 T9 R1 T5 R13
D7 /F  / C7/E / F 7/E / BΔ /D / Gm7 5 / C7

T9 R12 T5 R123 T5 R123 T5 R12 T5 R3

C7 / Am7 / D7 / Gm7 / C7 / F Δ /A

In Giant Steps of John Coltrane, the dance of moves T3 R3 or T5 R3

leads each time to the moves T6 R123 followed by T5 R12 , except at
the end where the transposition at the tritone is replaced by the
transposition a tone lower T10 R123 /T5 R12 . This creates a kind of
stability that is well highlighted by the Rameau–Schillinger operator
analysis:
T3 R3 T5 R3 T3 R3 T5 R3 T6 R123
BΔ / D7 / GΔ / B7 / EΔ / Am7 T5 R12 / D7

T5 R3 T3 R3 T5 R3 T3 R3 T5 R3 T6 R123 T5 R12
D7 / GΔ / B7 / EΔ / F 7 / BΔ / F m7 / B7

T5 R3
B7 / EΔ T6 R123/ Am7 T5 R12 / D7 T5 R3
/ GΔ T6 R123
/ Cm7 T5 R12 / F 7

T5 R3
F 7 / BΔ T6 R3 123/ F m T5 R12
/ B7 T5 R3
/ EΔ T10 R123/ C7 T5 R12
/ F 7

Definition 4.18. The elementary Abelian group E4 has order 4 and

presentation:
 
E4 = R1 , R2 | R12 = R22 = (R1 R2 )2 = 1
Definition 4.19. The seventh chords group RD3 = E4 × D12 is
defined as the product of the elementary Abelian group E4 by the
dihedral group D12 :

 
R1 , R2 , T, I | R12 = R22 = I 2 = T 12 = (T I)2
RD3 = E4 × D12 = = (R1 R2 )2 = R1 T R1 T −1 = R2 T R2 T −1
= (R1 I)2 = (R2 I)2 = 1
 Voice Leading and Neo-Riemannian Transformations 103

Theorem 4.15. The seventh chords group E4 ×D12 is a non-Abelian

group of order 96, isomorphic to C2 × C2 × D12 . The group does not

act transitively on U.

Proof. Since In Im = Im In , the group is not Abelian. Its order is

computed through the order of the product of two well-known groups:
|RD3 | = |E4 | × |D12 | = 4 × 24 = 96. As the group has fixed points,

it does not act transitively on U. 

The action of the group on the set U leads to 6 × 12 + 3 = 75

seventh chords and includes the transpositions of the 3-note chords
048 (444). Since by action of the dihedral group (which is isomorphic
to the T /I group), dominant seventh chords (433) are transformed
into half-diminished seventh chords (334) and minor major seventh
chords (344) are transformed into augmented major seventh chords
(443). The action on the basis rectangle of the eight types of the set
U is shown in Fig. 4.5.
The following example shows how the chords evolve in Fig. 4.5.
Starting from the upper-right corner (343), the chord Gm7 goes (R12 )
to the major seventh E7 (443), passes through the inversion I1 and
reaches the chord Bm75 of structure 334. Then the chord moves up
by R2 to the chord EmM 7 (344) crosses the T /I structure by the
transposition I0 in AΔ and finally returns to the upper-right corner
by R1 :

T9 R12 I1
Gm7 / E7 / Bm75 T5 R2 / EmM 7 I0
/ AΔ T3 R1 / Cm7

Fig. 4.5. 
Action of the group RD3 on the set U.
104 A Compendium of Musical Mathematics

4.9 Hierarchy of Rameau Groups

The generalization of the previous results to the ninth chords is

obvious. The ninth chords group RC4 = E16 × C12 is the product
of the elementary Abelian group E16 by the cyclic group C12 . RC4
has order 16 × 12 = 192. The ninth chords group RD4 = E8 × D12
is the product of the elementary Abelian group E8 by the dihedral
group. RD4 has order 8 × 24 = 192.
Seventh chords groups and ninth chords groups can be presented
more broadly by introducing the notion of Rameau groups or third
chain transformations. As we have seen in particular cases, there
are two ways to define Rameau Groups: as a product of the cyclic
group or as a product of the dihedral group. In both case, we intro-
duce the elementary Abelian group E2n of order 2n . The group has
presentation:
 
E2n = R1 , . . . , Rn | (Ri Rj )2 = 1, i, j = 1, . . . , n
(1) The Rameau group RCn is the product of the elementary Abelian
group E2n of order 2n by the cyclic group C12 :
 
R1 , . . . , Rn , T | (Ri Rj )2 = T 12 =
RCn = E2n × C12 =
T −1 Ri T Ri = 1, i, j = 1, . . . , n
It acts on the set of n-note chords built with major or minor thirds:
(words built with n letters on the alphabet {3, 4}, this could be
interesting for an application of these groups to rhythmic structures).
RCn has order |RCn | = 2n × 12. The groups are included in each
other:
RC1 ⊂ RC2 ⊂ RC3 ⊂ RC4 ⊂ · · ·
(2) The Rameau group RDn is the product of the elementary
Abelian group E2n−1 of order 2n−1 with the dihedral group D12 :

 
R1 , . . . , Rn−1 , T, I | (Ri Rj )2 = I 2 = T 12
RDn = E2n−1 × D12 = = Ri T Ri T −1 = (Ri I)2
= (T I)2 , i, j = 1, . . . , n − 1
The group has order |RDn | = 2n−1 × 24 = 2n × 12 = |RCn |. The
groups are included in each other:
RD1 ⊂ RD2 ⊂ RD3 ⊂ RD4 ⊂ · · ·
 Voice Leading and Neo-Riemannian Transformations 105

4.10 Exercises

Exercise 4.1
In 1880, Hugo Riemann presented two kinds of operations in the
tonnetz, Schritt and Wechsel. The Schritt Sn transposes a major
triad up by n semitones or a minor triad down by the same amount.
S0 is the identity. By indexing the major triad by capital M and
minor triad by m, Sn is defined by its values on the triads: S1 (CM ) =

CM , S1 (Cm ) = Bm , S2 (CM ) = DM , . . . , S11 (CM ) = BM , S11 (Cm ) =

Cm . The Wechsel Wn first transposes a major triad up and a minor
triad down by n semitones and then converts it to a triad of opposite
mode on the same root. Thus, Wn = P Sn , with P the usual neo-

Riemannian transformation: W0 (CM ) = Cm , W1 (CM ) = Cm ,. . . ,
W11 (CM ) = Bm . (1) Verify that W0 = P , W4 = L and W9 = R. (2)
For all triad X, show that (indices are considered mod 12)
Si Sj (X) = Si+j (X)
Si Wj (X) = Wj−i (X)
Wi Sj (X) = Wj+i (X)
Wi Wj (X) = Sj−i (X)
Exercise 4.2
Compare the action of the T /I subgroup {T0 , T4 , T8 , I3 , I7 , I11 } on
the set {CM , EM , AbM , Cm , Em , Abm } with the action of the S/W
subgroup {S0 , S4 , S8 , W0 , W4 , W8 } on the same set.
Exercise 4.3
Assign order numbers to the 24 triads such that C major is 1, C
major is 2, D major is 3, . . . , B major is 12, and C minor is 13,
C minor is 14, . . . , B minor is 24. Consider the permutation of the
Schritt
S = (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12)
(24, 23, 22, 21, 20, 19, 18, 17, 16, 15, 14, 13)
and the Wechsel
W = (1, 13)(2, 14)(3, 15)(4, 16)(5, 17)(6, 18)(7, 19)
(8, 20)(9, 21)(10, 22)(11, 23)(12, 24)
106 A Compendium of Musical Mathematics

(1) Show that W = W0 , S = S1 and Wn = W S n . (2) Show that

P = W , L = W S 4 and R = W S 9 . (3) Using the Schritt/Wechsel
relations, prove that P = (RL)3 R.
Exercise 4.4
Label the transformations of the following graph
DM / Bbm

 
F m / Dm

using P LR transformations and JQZ transformations.

Exercise 4.5
Let U and V be two transformations acting on the set of triads and
defined by
U (xM ) = (x + 1)M U (xm ) = (x + 1)m
V (xM ) = (x)m V (xm ) = (x + 6)M
Prove that U and V commute (U V = V U ). Assign the order
defined at Exercise 4.3, write U and V as permutations and recover
the relation U V = V U .
Exercise 4.6
Consider the presentation of the quaternion group
 
Q8 = u, v | u4 = 1, u2 = v 2 , uvu = v
Prove that Q8 is generated by u and v defined on triads by
u(xM ) = (x + 6)M v(xM ) = (3 − x)m
u(xm ) = (x + 6)m v(xm ) = (3 − x)M

References

Arnett, J. and Barth, E. (2011). Generalizations of the Tonnetz: Tonality

revisited. In: Proceedings of the 2011 Midstates Conference on Undergrad-
uate Research in Computer Music and Mathematics.
Berry, C. and Fiore, T. (2018). Hexatonic systems and dual groups in mathemat-
ical music theory, Involve a Journal of Mathematics 11(2), 253–270.
 Voice Leading and Neo-Riemannian Transformations 107

Cannas, S. (2018). Geometric Representation and Algebraic Formalization of

Musical Structures, PhD. Dissertation, University of Strasbourg and
Università degli Studi di Pavia e di Milano-Biocca.
Cannas, S., Antonini, S. and Pernazza, L. (2017). On the group of transformations
of classical types of seventh chords. In: Agustı́n-Aquino O., Lluis-Puebla, E.
and Montiel, M. (eds.) Mathematics and Computation in Music, Lecture
Notes in Computer Science, Vol. 10527, Springer, Mexico City, 13–25.
Childs, A. (1998). Moving beyond neo-Riemannian triads: Exploring a transfor-
mational model for seventh chords, Journal of Music Theory 42(2), 191–193.
Cohn, R. (1997). Neo-Riemannian operations, parsimonious trichords, and their
Tonnetz representation, Journal of Music Theory 41(1), 1–66.
Cohn, R. (1998). Introduction to neo-Riemannian theory: A survey and a histor-
ical perspective, Journal of Music Theory 42(2), 167–180.
Cohn, R. (2012). Audacious Euphony. Chromaticism and the Triad’s Second
Nature, Oxford University Press, New York.
Crans, A., Fiore, T. and Satyendra, R. (2009). Musical actions of dihedral groups,
The American Mathematical Monthly 116(6), 479–495.
Douthett, J. and Steinbach, P. (1998). Parsimonious graphs: A study in par-
simony, contextual transformation, and modes of limited transposition,
Journal of Music Theory 42(2), 241–263.
Fiore, T. and Noll, T. (2011). Commuting groups and the topos of triads.
In: Agon, C. Andreatta, M., Assayag, G., Amiot, E., Bresson, J. and
Mandereau, J. (eds.) Mathematics and Computation in Music. MCM 2011.
Lecture Notes in Computer Science, Vol. 6726, Springer, Berlin, 69–83.
Fiore, T. and Noll, T. (2018). Voicing transformations and a linear representation
of uniform triadic transformations, Siam Journal on Applied Algebra and
Geometry 2(2), 281–313.
Fiore, T., Noll, T. and Satyendra, R. (2013). Morphisms of generalized interval
systems and PR-groups, Journal of Mathematics and Music 3, 3–27.
Fiore, T. and Satyendra, R. (2005). Generalized contextual groups, Music Theory
Online 11(3).
Gollin, E. (1998). Some aspects of three-dimensional Tonnetze, Journal of Music
Theory 42(2), 195–206.
Hook, J. (2002). Uniform triadic transformations, Journal of Music Theory
46(1/2), 57–126.
Hyer, B. (1989). Tonal Intuitions in Tristan Und Isolde, PhD dissertation, Yale
University.
Hyer, B. (1995). Reimag(in)ing Riemann, Journal of Music Theory 39(1),
101–138.
Jedrzejewski, F. (2005). Permutation groups and chord tesselations. In: Inter-
national Computer Music Conference, ICMC 2005, Barcelona, Spain,
September 4–10, 2005, Michigan Publishing.
Jedrzejewski, F. (2019a). Non-contextual JQZ transformations. In: Montiel, M.,
Gomez-Martin, F. and Agustı́n-Aquino, O.A. (eds.) Mathematics and Com-
putation in Music, 7th International Conference, MCM 2019, Madrid,
Spain, June 18–21, 149–160.
108 A Compendium of Musical Mathematics

Jedrzejewski, F. (2019b). The hierarchy of Rameau groups. In: Montiel, M.,

Gomez-Martin, F. and Agustı́n-Aquino, O.A. (eds.) Mathematics and Com-
putation in Music, 7th International Conference, MCM 2019, Madrid,
Spain, June 18–21, 161–171.
Kerkez, B. (2012). Extension of neo-Riemannian PLR-group to seventh chords,
Bridges 2012, Mathematics, Music, Art, Architecture, Culture, 485–488.
Klumpenhouwer, H. (1991). Aspects of row structure and harmony in Martino’s
impromptu number 6, Perspectives of New Music 29(2), 318–354.
Lewin, D. (1982). A formal theory of generalized tonal functions, Journal of Music
Theory 26, 23–60.
Lewin, D. (1990). Klumpenhouwer networks and some isographies that involve
them, Music Theory Spectrum 12(1), 83–120.
Popoff, A. (2013). Building generalized neo-Riemannian groups on musical trans-
formations as extensions, Journal of Mathematics and Music 7(1), 55–72.
Riemann, H. (1887). Handbuch der Harmonielehre, Leipzig.
Tymoczko, D. (2011). A Geometry of Music. Harmony and Counterpoint in the
Extended Common Practice, Oxford University Press, New York.
Waller, D.A. (1978). Some combinatorial aspects of the musical chords, The Math-
ematical Gazette 62, 12–15.
 Chapter 5

Combinatorics on Musical Words

Classification of modes has been studied by David Clampitt and

Thomas Noll (2018). The aim of this chapter is to study a new
classification of modes based on plactic congruences. These congru-
ences mimic a small perturbation from one mode to the other by the
move of only one note. Two modes are in the same plactic class if
they are related by a path of modes which are pairwise linked by
plactic congruences. A mode is an ordered series of musical inter-
vals (or steps). A scale is an ascending or descending series of notes,
representing a class of modes under circular permutations. In tra-
ditional Western music, the C major scale represents the circular
permutations of the seven usual modern modes (Ionian, Dorian,
Phrygian, etc.). In the first section, we review rudiments of formal
languages (see Berstel et al., 2008; Lothaire, 2002). The use of for-
mal languages is also a new way of considering the problems of voice
leading and neo-Riemannian transformations.

5.1 Musical Words

Definition 5.1. Let A be a finite set of symbols, called the alphabet.

The elements of A are called letters. A word ω is written uniquely
by ω = a1 · · · ar with letters aj ∈ A for j = 1, . . . , r. The identity
element ε (or 1) is called the empty word. The set of words is denoted
by A∗ .

Definition 5.2. The concatenation uv of two words u = u1 · · · un

and v = v1 · · · vm is the word uv = u1 · · · un v1 · · · vm .
109
110 A Compendium of Musical Mathematics

Definition 5.3. A monoid is a set equipped with an associative

binary operation and has a neutral element.
Definition 5.4. The free monoid A∗ on a set A is the monoid whose
elements are all finite sequences (or strings) of zero or more elements
from A, equipped with string concatenation and the empty element
as the identity element.
Definition 5.5. The set A+ = A∗ \{ε} is called the free semigroup
over A.
Definition 5.6. A word ω is called a factor (resp., a prefix, resp., a
suffix ) of a word u if there exist words x, y such that u = xωy (resp.,
u = ωy, resp., v = xω). The factor is proper if xy = ε.
Definition 5.7. The length of ω = a1 · · · ar , with letters aj ∈ A is r
and denoted by |ω|. If ω ∈ A∗ and a ∈ A is a letter, |ω|a denotes the
number of occurrences of the letter a in the word ω:

|ω| = |ω|a
a∈A

The lexicographic order (also called the alphabetic order) is

defined as follows.
Definition 5.8. Let x, y be two words of A∗ . We said that x < y if
x is a proper prefix of y or if there exist factorizations x = uax , and
y = uby  with a, b letters of A and a < b.
 of a word ω = a1 · · · ar is the word
Definition 5.9. The reverse ω
 = ar ar−1 · · · a1 .
ω
Definition 5.10. A word ω ∈ A∗ is called primitive if it is not a
power of another word.
Theorem 5.1. A primitive word of length n has exactly n distinct
cyclic permutation.
Theorem 5.2. A word ω ∈ A∗ is primitive if and only if ω = un
with n ≥ 0 implies ω = u.
Corollary 5.1. If two words u and v commute (uv = vu), then there
are power of a same word x.
 Combinatorics on Musical Words 111

Definition 5.11. Two words u and v are conjugate if there exist

words x, y such that u = xy and v = yx.
Let σ : A∗ → A∗ defined by σ(1) = 1 and σ(au) = ua for a ∈ A,
u ∈ A∗ .
Theorem 5.3. Two words u and v are conjugate if and only if there
exists an integer n ≥ 0 such that v = σ n (u).
Definition 5.12. If ω ∈ A∗ , and if the alphabet A is a set of integers,
the height h(ω) of ω = a0 a1 · · · ar−1 is the sum of its letters

r−1
h(ω) = aj = a0 + a1 + · · · + ar−1
j=0

Formal words are used to represent either classes of pitches, scales,

chords, rhythms or any musical objects. In the representation of
pcsets, it suffices to consider the alphabet A = ZN . For instance,
a musical scale in the N -EDO can be coded by a sequence of inte-
gers in which each term is the number of steps within consecutive
notes of the scale. For instance, for N = 12, the major scale is coded
2212221. In the representation of rhythms, we will assume there
is some basic, invariant unit pulse that cannot be divided. Several
models are available for rhythms representation. For instance, one
can consider the representation of durations or the representation of
onsets. In the latter, a rhythm pattern is a sequence of note onsets
and two rhythm patterns are the same if they have the same sequence
of onsets. In the former (that we subsequently adopt), each letter cor-
responds to the duration of a note or rest. If necessary, the distinction
between pitches and rests is made by considering the alphabet A∪A,
with A representing pitches and A rests. For instance, the 322 over
{2, 3}∪{2, 3} represents a note of 3 units, followed by a rest of 2 units
and a note of 2 units.

5.2 Syntactic Monoids

Definition 5.13. A morphism ϕ : M → N from a monoid M to a

monoid N satisfies for all x, y ∈ M
ϕ(xy) = ϕ(x)ϕ(y)
and furthermore ϕ(1M ) = 1N .
112 A Compendium of Musical Mathematics

Definition 5.14. Let N be a monoid. A subset X ⊆ N is recognized

by a monoid M if there exists a morphism ϕ : M → N such that
X = ϕ−1 (ϕ(X)). A subset is said recognizable if it is recognized by
some finite monoid.
Definition 5.15. Let X be a subset of a free monoid M . The syn-
tactic equivalence x ∼X y is defined by

∀u, v ∈ M, uxv ∈ X ⇔ uyv ∈ X

The quotient monoid M (X) = M/∼X is called the syntactic monoid

of X.
Proposition 5.1. The syntactic monoid is the smallest monoid that
recognizes X.
Proposition 5.2. Let M be a monoid. A subset X of M is recog-
nizable if and only if its syntactic monoid M (X) is finite.
Proposition 5.3. The set Rec(M ) of recognizable subsets of M
is closed under union, intersection, complement and right and left
quotient.
Remark 5.1. Rec(M ) is not closed under product and under Kleene
star. For instance, X = {(1, 1)} is a recognizable subset of N2 . But
X ∗ = {(n, n), n ∈ N} is not recognizable since its syntactic monoid
is infinite.
Definition 5.16. Let M be a monoid. The set Rat(M ) of rational
subsets of M is the smallest set that contains every finite set and is
closed under union, product and Kleene star.
Remark 5.2. Rat(M ) is not closed under complement.
Theorem 5.4 (Kleene, 1936). The recognizable languages Rec(M )
coincide with the rational languages Rat(M )

Rec(M ) = Rat(M )

Definition 5.17. The set of star-free languages is the smallest set

containing all finite languages and closed by union, concatenation
and complement.
Example 5.1. The language L = (ab)∗ is star-free.
 Combinatorics on Musical Words 113

Definition 5.18. A finite monoid M is aperiodic if there exists n ≥ 0

such that for all x ∈ M , xn = xn+1 .
Definition 5.19. A language is aperiodic if it could be recognized
by an aperiodic monoid.
Theorem 5.5 (Schützenberger). A language is star-free if and
only if its syntactic monoid is aperiodic.
Example 5.2. The language L = (aa)∗ is not star-free.

5.3 Formal Grammars

Definition 5.20. A grammar G = (T, N, P, S) consists of the fol-

lowing components: (1) a finite set T of terminal symbols (denoted
by lowercase letters), (2) a finite set N of non terminal symbols
(denoted by capital letters) and disjoint from T and (3) a finite set
P of production rules of the form u → v with u and v are sequences
of symbols of N ∪ T
(T ∪ N )∗ N (T ∪ N )∗ → (T ∪ N )∗
(4) A start symbol S ∈ N, called axiom. A grammar G generates a
language denoted L(G).
Example 5.3. Let N = {S, B}, T = {a, b}, S → Ab, S → 1, A →
aS, B → b. This grammar defines the language
L(G) = {an bn , n ≥ 1}
Formal grammars have been used in music theory since the
1980s. Marc Steedman (1984), Marc Chemillier (2004), Fred Lerd-
hal and Ray Jackendoff (1996) intensively developed new forms of
analysis, sometimes inspired by the Schenkerian theory. French com-
poser Philippe Manoury (2012) applied formal grammars to his own
compositions.
Definition 5.21. A Lindenmayer system (or L-system) (V, S, P ) is
a formal grammars consisting of a set V of symbols that can be
replaced (variables) or not (terminals), an axiom S which is a string
of symbols from V defining the initial state and P is a set of produc-
tion rules.
114 A Compendium of Musical Mathematics

Example 5.4. (1) The L-system V = {a, b}, S = a, P : a → b, b →

ab defines Fibonacci’s words: a, b, ab, bab, abbab, etc. (2) The L-system
V = {a, b}, S = a, P : a → aba, b → bbb defines Cantor’s set.
Spanish composer Alberto Posadas intensively used L-system in
his string quartets Liturgia Fractal (2003–2007), Oscuro abismo de
llanto y de ternura (2005) and Glossopoeia (2009). His works have
been studied by musicologist José Luis Besada (2017).

5.4 Words and Rhythms

5.4.1 Lyndon words

Some rhythms are Lyndon words.
Definition 5.22. A Lyndon word is a primitive word ω ∈ A+ that is
less than all its conjugate in the lexicographic order. In other words,
ω is a Lyndon word if and only if for all u, v ∈ A+ such that ω = uv,
we have ω < v.
If σk (ω) denotes the circular permutation of ω at position k, which
is the conjugate of ω at order k, we have ω < σk (ω) for all k =
1, 2, . . . , |ω|.
Example 5.5. The first Lyndon words on {a, b} are a, b, ab,
aab, abb, aaab, aabb, abbb, etc.
Proposition 5.4. If u, v are Lyndon words with u < v, then uv is a
Lyndon word.
Example 5.6. aab, ab are Lyndon words with aab < ab, thus aabab
is a Lyndon word.
Proposition 5.5. The product of two Lyndon words u, v is a Lyndon
word if and only if u <lex v.
Theorem 5.6. Any word ω factorizes uniquely as a nonincreasing
product of Lyndon words lj
ω = l1n1 l2n2 · · · lrnr
with l1 > l2 > · · · > lr and nj ≥ 1. This factorization is called the
Lyndon factorization.
 Combinatorics on Musical Words 115

Example 5.7. The word ω = aabaabba = (aab)2 (a).

Proposition 5.6. The number of Lyndon words of length n over k

letters is
1
μ(d)kn/d
n
d|n

where μ is the Möbius function.

5.4.2 Euclidean rhythms

Definition 5.23. A word ω = a1 · · · an of positive integers is a

Euclidean string if increasing a1 by 1 and decreasing an by 1 yield a
new word

ω  = (a1 + 1)a2 · · · an−1 (an − 1)

which is a rotation of ω.

Example 5.8. Cuban rhythm cinquillo ω = 12122 is a Euclidean

string since ω  = 22121 is a rotation of ω. Rhythms fume-fume 22323,
aksak 223, bembé 1221222, etc. are Euclidean strings. The word 332
(which leads to 431) is not Euclidean, but its rotation 233 (ω  = 332)
is Euclidean.

John Ellis and his collaborators (Ellis et al., 2003) demonstrated

the following results:

Theorem 5.7. A word ω = a1 · · · an is a Euclidean string if and

only if n and h(ω) = a1 + · · · + an = k are coprime, and in this case,
the string is unique.

Theorem 5.8. Euclidean strings are Lyndon words.

The algorithm to generate a Euclidean chain E(n, k) of length

n and height k consists in introducing two transformations a and b
acting on each letter x of ω as follows:

a(x) = x + 1, b(x) = 01x

116 A Compendium of Musical Mathematics

where 1x is the symbol 1 repeated x time. Using these two applica-

tions, we build the chain by induction:
⎧
⎨ aE(n, k − n)
⎪ si n < k
E(n, k) = bE(n − k, k) si n > k
⎪
⎩
1 si n = k

Example 5.9. For instance, the Euclidean string E(3, 8) is the

Cuban tresillo rhythm:

E(3, 8) = aE(3, 5) = aaE(3, 2) = aabE(1, 2) = aabaE(1, 1)

= aaba(1) = aab(2) = aa(011) = a(122) = (233)

Theorem 5.9. The word ω = ω0 ω1 · · · ωn−1 where each term is

given by

nj
ωi = card j : n − − 1 = i, 0 ≤ j < k
k

is a Euclidean string E(n, k) of length n and height k, where x is

the floor function of x.

5.4.3 Maximally even rhythms

John Clough and Jack Douthett introduced in 1991 the notion of
maximally even scales. This notion is easily transposed to rhythms.
Definition 5.24. The conjugates of a word ω = ω0 ω1 · · · ωn−1 of
length n and height h(ω) = ω0 + ω1 + · · · + ωn−1 are the words ω (j)
(j)
with j = 1, . . . , n − 2 of length n such that each letter ωi is the sum
of j consecutive symbols counted circularly modulo the height of ω:
(j)
ωi = ωi + ωi+1 + · · · + ωi+j−1 mod n mod h(ω)

Definition 5.25. A word is maximally even if each conjugate takes

at least two consecutive values.
Theorem 5.10. Euclidean strings are maximally even.
Example 5.10. Consider the bossa nova rhythm represented by
the word ω = 33433 of length n = 5 and height 16. For each
 Combinatorics on Musical Words 117

value of j, form the conjugate word. Thus, we have for j = 1,

ω (1) = (6, 7, 7, 6, 6), for j = 2, ω (2) = (10, 10, 10, 9, 9) and for j = 3,
ω (3) = (13, 13, 13, 12, 13), which shows that the word 33433 is max-
imally even since ω (3) comes into two consecutive values. The souk-
ous rhythms (Congo) 33415, gahu (Ghana) 33442 and rumba 34324
(Latin America) are not maximally even because they have three
values from the first conjugate.

5.4.4 Deep rhythms

Deepness is a property of rhythms and scales which relates to
the number of occurrences of the distances defined by pairs of
onsets. Consider a rhythm of length n (with n onsets) and height
k represented as a set of n points on a circle of circumference k.
We define the multiplicity of every distance between all pairs of onsets
as the number of occurrences. The notion of deepness was introduced
by Terry Winograd (1966) (for scales) and popularized by Carlton
Gamer (1967). It has been extensively used by composer Tom John-
son in various scores, such as Deep Rhythms for three Performers
(2019).
Definition 5.26. A rhythm represented by a word ω = a1 · · · an
of length n and of height a1 + · · · + an = k is Winograd-deep (or
W-deep) if every distance 1, 2, . . . , k/2 has a unique multiplicity.
Example 5.11. The rhythm [×××·×·] or {0, 1, 2, 4}6 represented by
the word 1122 is W-deep because distance 1 appears twice, distance
2 appears thrice and distance 3 appears once.
Remark 5.3. Equivalently, a scale is Winograd-deep if the number
of pitches it has in common with each of its rotations is unique. This
equivalence is the Common Tone Theorem (see Chapter 2: p. 39).
Paul Erdős introduced another definition of deep rhythms in
Erdős (1989).
Definition 5.27. A rhythm with n onsets is Erdős-deep (or E-deep)
if for every multiplicity i = 1, 2, . . . , n − 1, there is a nonzero distance
with exactly that multiplicity.
In other words, a rhythm is Winograd-deep if all distances
between 1 and n − 1 appear a unique number of times. While in
118 A Compendium of Musical Mathematics

a Erdős-deep rhythm, each (nonzero) distance that appears has a

unique multiplicity.

Proposition 5.7. Every Winograd-deep rhythm is also Erdős-deep.

Example 5.12. The converse is false. For instance, the bossa nova
rhythm 33433 which is written on the circle {0, 3, 6, 10, 13}16 has the
following distances

0 3 6 10 13
0 0 3 6 6 3
3 3 0 3 7 6
6 6 3 0 4 7
10 6 7 4 0 3
13 3 6 7 3 0

and occurrences

d 1 2 3 4 5 6 7
i 0 0 4 1 0 3 2

The bossa nova rhythm is Erdős-deep but not Winograd-deep

because distances 1, 2 and 5 do not appear.

Perouz Taslakian gave in his thesis a characterization of E-deep

rhythms (Taslakian, 2008). He demonstrated the following results.

Theorem 5.11 (Taslakian, 2008). A rhythm is Erdős-deep if and

only if it is a rotation of a scaling of either the rhythm {0, 1, 2, 4}6 or
the rhythm {jm mod n, j = 0, 1, . . . , (k − 1)} for some k, n, m with
k ≤ n/2g + 1 and g = gcd(m, n).

Example 5.13. The bossa nova rhythm {0, 3, 6, 10, 13}16 is E-deep;
it is the set of multiples for m = 13, starting at 6, {6, 3, 0, 13, 10}.
The rhythm {0, 4, 5, 9, 10, 14, 15}16 is W-deep and thus E-deep. It is
the set of multiples of m = 5, {0, 5, 10, 15, 4, 9, 14}.

Theorem 5.12 (Taslakian, 2008). A maximally even rhythm sat-

isfying k ≤ n/2 + 1 is E-deep if and only if k is coprime with n.
 Combinatorics on Musical Words 119

5.4.5 Rhythmic oddity

The rhythmic oddity property was discovered by ethnomusicologist
Simha Aron (1991) in the study of Aka pygmies music. The rhythms
satisfying this property are refinement of Aksak rhythms described by
Brăiloiu (1952) and are also used by Turkish and Bulgarian music.
Turkish composer Ahmed Saygun (1907–1991) wrote many piano
pieces that use Aksak rhythm. The number of words satisfying the
rhythmic oddity property is given in Jedrzejewski (2017b).
Definition 5.28. A word ω defined on the alphabet {2, 3} satisfies
the rhythmic oddity property if its height h(ω) (the sum of its letters)
is even and no cyclic shift of ω can be factorized into two words uv
such that h(u) = h(v).
Example 5.14. The word 32322 of height 12 has the rhythmic odd-
ity property, as well as all words of the form 32n 32n+1 for all positive
integer n. The words satisfying the rhythmic oddity property is dis-
tinct from Lyndon words: the words 222 and 233233233 satisfy the
rhythmic oddity property but are not Lyndon words. Conversely, the
Lyndon word 2233 has not the rhythmic oddity property (the words
23 and 32 have the same height and 2332 is a rotation of 2233).
The cycle δ of ω is defined by δ(ε) = ε and δ(aω) = ωa, for
a ∈ {2, 3}. The rotations of ω are the words δi (ω), for all positive
integer i > 0. In 2010, André Bouchet (2010) gave the following
characterization.
Theorem 5.13. Let ω = ω0 ω1 · · · ωn−1 be a word over the alphabet
A = {2, 3} of height 2h. The word ω satisfies the rhythmic oddity
property if and only if the two conditions are satisfied:
(1) The length of ω is odd, say 2 + 1.
(2) The height of the prefixes of length  of the rotations δi (ω) of ω
is equal to h − 2 or h − 1.
The proof of this theorem is based on the two following properties.
Proposition 5.8. Let ω be a word of length n satisfying the rhythmic
oddity property and let i and λ be two integers such that 0 ≤ λ < n:
(1) If the height of the prefix of length λ of the word δi (ω) is h − 1
or h − 2, then the height of the prefix of δi+1 (ω) is h − 1 or h − 2.
(2) The word ω has a prefix of height h − 1 or h − 2.
120 A Compendium of Musical Mathematics

Proposition 5.9. If ω ∈ {2, 3}∗ of height 2h satisfies the rhythmic

oddity property, then

(1) the number |ω|2 of symbol 2 in ω is odd,

(2) the number |ω|3 of symbol 3 in ω is even,
(3) the length of ω is odd,
(4) δi (ω) satisfies the rhythmic oddity property for any i,
(5) ω and δi (ω) have a prefix of height h − 1 or h − 2.

Since the height of ω satisfies 2 |ω|2 + 3 |ω|3 = 2h, the number

of occurrences of letter 3 is even. Thus, since the length of ω is odd
by the previous proposition, the number of occurrences of letter 2 is
even.
Furthermore, André Bouchet defines d-pairing. Let ω be a word
of length n and d be an integer such that 0 < d ≤ n/2. A d-pairing
of ω is a partition of the subset of indices {i : 0 ≤ i < n, ωi = 3} into
pairs of indices {j, j + d}. Arithmetic operations on indices are to be
understood mod n. Let ω = ω0 ω1 · · · ωn−1 be a word of {2, 3}∗ and
d a positive integer coprime with n. Denote by ω (d) = x0 x1 · · · xn−1
the word obtained by reading all letters of ω by step d, starting from
ω0 . Namely, each letter of ω (d) is xj = ωk with k = jd mod d,
0 ≤ j < d. For instance, the word ω = 2233233 admits a 3-pairing
and ω (3) = 2333322. As a corollary of the previous result, Bouchet
shows the following result, which is the theoretical meaning of the
Hop-and-jump algorithm given by Godfried T. Toussaint (2013).

Theorem 5.14. Let ω be a word of even height. ω satisfies the rhyth-

mic oddity property if and only if the two conditions are satisfied:

(1) The length of ω is odd, say 2 + 1.

(2) ω admits a -pairing.

Some years before, Marc Chemillier and Charlotte Truchet (2003)

gave another characterization of rhythmic oddity property by intro-
ducing asymmetric pairs.

Definition 5.29. The words (u, v) form an asymmetric pair if no

pair of prefixes (u , v  ) of u and v respectively exist such that h(u ) =
h(v  ) + 1.
 Combinatorics on Musical Words 121

Example 5.15. For instance, (2233,233) is an asymmetric pair, but

(2232,333) is not since the pair of prefixes (22,3) verifies h(22) =
h(3) + 1.
Theorem 5.15. A word ω satisfies the rhythmic oddity property if
and only if there exists an asymmetric pair (u, v) such that ω = uv
or ω = vu with h(v) = h(u) + 2.
A construction of asymmetric pairs, given by Chemillier and
Truchet, uses two functions over the monoid A∗ × A∗ into itself,
namely
r(u, v) = (3u, 3v), s(u, v) = (v, 2u)
and identifies any word ω of A∗ with a word α over {r, s}∗ by the
map ω → f (ω) such that f (ω) = α(1, 1) where 1 is the empty word.
For instance, the Cuban tresillo rhythm represented by the word 332
is associated with the word rs, since (3, 32) = r(1, 2) = rs(1, 1). The
characterization of words ω with the rhythmic oddity property is
then moved to a property of the associated word α.
Theorem 5.16. A word ω satisfies the rhythmic oddity property if
and only if there exists a word α ∈ {r, s}∗ with |α|s being odd, such
that ω = uv or vu with (u, v) = α(1, 1).

5.5 Words and Scales

5.5.1 Dyck words

Dyck words play a crucial role in the theory of formal languages and
in its applications to music.
Definition 5.30. A Dyck word ω on alphabet A = {a, b} is a word
that contains has many as as bs |ω|a = |ω|b and such that all prefix
u of ω contains no more bs than as, |u|b ≤ |u|a .
Originally, Dyck words were defined as “well-parenthesized
words” over the alphabet Σ = {(, )}, that is, pairs of parentheses
which are correctly matched.
Definition 5.31. If for a word ω ∈ Σ∗ , we define imb(ω) = |ω|( −
|ω|) , the number of occurrences of the left parenthesis minus the
122 A Compendium of Musical Mathematics

number of occurrences of the right parenthesis. ω is a Dyck word if

and only if imb(ω) = 0 and imb(u) ≥ 0 for all prefix u of ω.
Another way of defining Dyck words is to consider the following
reduction.
Definition 5.32. Let A be an alphabet and A = {a, a ∈ A} a copy
of A disjunct of A. On the alphabet A∪A, the reduction is defined
by ω → ω  if there exists a factorization ω = xaay, a ∈ A such that
ω  = xy.
Definition 5.33. A Dyck word is a word that can be reduced to ε.
Example 5.16. Applying the reduction to the word ω =
abaabbba → abbbba → abba → aa → ε leads to the neutral element.
Definition 5.34. A (n, n)-Dyck path is a staircase walk on the inte-
ger lattice Z2 from (0, 0) to (n, n) that lies strictly above [or under]
(but may touch) the diagonal y = x.
Proposition 5.10. The number of Dyck paths (staircase from (0, 0)
to (n, n)) is the Catalan number
1 2n (2n)!
Cat(n) = =
n+1 n (n + 1)!n!

5.5.2 Infinite words and complexity

Generally, infinite words (at right) are given as the limit of the
sequence of finite words. The set of the infinite words over the alpha-
bet A is denoted by Aω .
Example 5.17. The Fibonacci words are defined recursively by ω0 =
a, ω1 = b, ωn = ωn−1 ωn−2
ω = lim ωn = babbababba....
Example 5.18. The Thue–Morse word is defined by the rewriting
rules a → ab, b → ba
ω = abbabaabbaab.....
Example 5.19. The Narayana words are defined by ω0 = a, ω1 = b,
ω2 = baa, ω3 = baaa, ωn+1 = ωn ωn−2 . This sequence has been used
by composer Tom Johnson in Les vaches de Narayana (1998).
 Combinatorics on Musical Words 123

Complexity of words measures the diversity of the subwords of a

given word.

Definition 5.35. The complexity of the infinity word ω is the num-

ber that counts for each integer n, the number σn (ω) of distinct
factors of length n in ω.

Example 5.20. The complexity of the Fibonacci word ω is σn (ω) =

n + 1.

5.5.3 Sturmian words

Definition 5.36. A Sturmian word ω on a binary alphabet is an

infinite word such that its complexity is σn (ω) = n + 1. In other
words, the number of distinct factors of ω of length n is n + 1.
On an alphabet of size k ≥ 2, a Sturmian word is an infinite word of
complexity n + k − 1.

Example 5.21. The Fibonacci word is Sturmian.

Definition 5.37. A word ω is k-balanced if for any letter a of the

alphabet A and for all distinct factors u, v of ω of the same length
n = 1, 2, . . . , |ω|, the number of occurrences of letter a in u and v
differ by at most k + 1 values:

|u| = |v| =⇒ ||u|a − |v|a | ≤ k

A word ω is a balanced word (= a 1-balanced word) if for any letter

a of the alphabet A and for all factors u, v of ω of the same length
n = 1, 2, . . . , |ω|, the number of letters a in u and in v differs by at
most 1.

Definition 5.38. An infinite word ω is periodic if for some positive

integer p and for all positive integer n, ωn+p = ωn . In this case, ω
can be written as ω = uuuuu · · · = uω with u of length p. The word
ω is ultimately periodic if it can be written as ω = uv, with u ∈ A∗
(finite) and v ∈ Aω , v infinite periodic.

In other words, ω is periodic from a certain index. The infinite

word ω is aperiodic if it is not ultimately periodic, that is, it does
not consist of a finite sequence followed by a finite cycle.
124 A Compendium of Musical Mathematics

Proposition 5.11. A finite word is a factor of some Sturmian word

if and only if it is balanced.

Theorem 5.17 (Morse and Hedlund, 1940). An infinite word

ω on a binary alphabet is Sturmian if and only if ω is balanced and
aperiodic.

The proof is based on two propositions. If ω is not ultimately

periodic, then σn (ω) ≤ n + 1. And if ω is balanced, then σn (ω) ≥
n + 1.
Sturmian words can be build using irrational rotation. Let α ∈
[0, 1) be an irrational number. Consider the rotation defined by Tα :
[0, 1) → [0, 1), x → x + α mod 1. Given a starting point x0 ∈ [0, 1),
compute the sequence

a if Tαn (ω) ∈ [0, α)
ωn =
b otherwise

Then the word ω = ω1 ω2 · · · is a Sturmian word on {a, b}. The con-

verse is also true. Given a Sturmian word, one can find two generators
α, ρ. The proof is given in Justin and Pirillo (1997). More precisely,
we have the following theorem.

Theorem 5.18. Let (ωn ) be a Sturmian sequence on {a, b}. Then

there exists unique α, ρ ∈ [0, 1) with α irrational, such that ωn is
given by either

a if (n + 1)α + ρ − nα + ρ = 0
ωn =
b otherwise

or

a if (n + 1)α + ρ − nα + ρ = 0
ωn =
b otherwise

for all n, where x is the floor function (greatest integer ≤ x) and

x is the ceiling function (the smallest integer ≥ x).

Theorem 5.19. A word is Sturmian if and only if it is generated by

an irrational rotation.
 Combinatorics on Musical Words 125

5.5.4 Christoffel words

Let ω be a finite word on an alphabet {a, b}. Consider the slope
s/t, where s is the number of occurrences of a in ω and t is the
number of occurrences of b in ω. The word ω can be drawn on a
grid representing a as a horizontal unit segment and b as a vertical
unit segment (see Fig. 5.1). The line D connects the origin (0, 0) to
the point (s, t), dividing the grid in an upper part and a lower part.
A Christoffel word is the discretization of a line segment by a path
in an integer lattice. More precisely, the (lower) Christoffel word (see
Reutenauer, 2019) is the word that lies strictly under (or above) the
diagonal D (but may touch) while staying as close as possible to the
diagonal.
For instance, the Christoffel word of Fig. 5.1 is

C(8/5) = aabaababaabab

The upper Christoffel path is the mirror image of the lower Christoffel
path. Looking at squares crossed by the line D, the discrete path
joining the center of each square encodes the billiard word:

B(8/5) = xyxxyxyxxyxy

There is a bijection between Christoffel words and billiard words.

Proposition 5.12 (Arnoux). The complexity of a billiard word is
σ(n) = n2 + n + 1.

Fig. 5.1. Christoffel and billiard words.

126 A Compendium of Musical Mathematics

Proposition 5.13. Let C(α) be a Christoffel word of slope α. Then

C(α) = aωb, where ω is a palindrome.
Example 5.22. For instance, the billiard word ω = B(8/5) is a
palindrome and C(8/5) = aωb.
Theorem 5.20. A word ω is a Christoffel word if and only if it is
a balanced Lyndon word.
Definition 5.39. The standard factorization of the Christoffel word
ω of slope b/a is the factorization ω = (ω1 , ω2 ), where ω1 encodes the
portion of the Christoffel path from (0, 0) to the closest point P to
the diagonal line segment D and ω2 encodes the portion from P to
(a, b).

Theorem 5.21 (Borel–Laubie, 1993). A proper Christoffel word

ω has a unique factorization (called the standard factorization) ω =
(u, v), where u and v are Christoffel words.

Example 5.23. The Christoffel word C(8/5) factorizes C(8/5) =

(u, v) = (aabaabab, aabab), where u = C(5/3) is the Christoffel word
of slope 5/3 and v = C(3/2) is the Christoffel word of slope 3/2.
The standard factorization of Christoffel words leads to the palin-
dromic factorization.
Theorem 5.22. Let ω be a Christoffel word, with ω = (aub)(avb) its
standard factorization. The palindromic factorization is the unique
factorization ω = ω1 ω2 , where ω1 = ava and ω2 = bub.
Proof. Since ω = (aub)(avb) is a Christoffel word, the word
ω = ubav = vabu is a palindrome. Thus, ω = ω1 ω2 = a(vabu)b =
a(ubav)b. 

Example 5.24. The Christoffel word C(8/5) = (aub)(avb) with u =

abaaba and v = aba has a palindromic factorization ω1 = a(aba)a and
ω2 = bub = b(abaaba)b. Remark that the palindromic factorization
is different from the standard factorization.

5.5.5 Sturmian morphisms

In this section, all words are defined on a two-letter alphabet {a, b}.
 Combinatorics on Musical Words 127

Diatonic modes and their transpositions are described by the

Douthett function. This function represents the scales of k notes in
N -EDO:

m Nx + m
JN,k (x) = mod N
k

In the 12-EDO, there exist 84 = 12 × 7 ways of transposing the 7

diatonic modes twelve times. These 84 scales of 7 pitches (k = 7) are
obtained for m ∈ Z12 :
0
J12,7 = {0, 1, 3, 5, 6, 8, 10} = locrian on C
1
J12,7 = {0, 1, 3, 5, 7, 8, 10} = phrygian on C (5.1)
···
83
J12,7 = {11, 1, 3, 5, 6, 8, 10} = lydian on B (5.2)

David Clampitt and Thomas Noll (see Clampitt and Noll, 2011;
Noll, 2009) have developed a theory of diatonic modes based on the
Sturmian monoid that we now define.
Definition 5.40. A morphism f is Sturmian if for every Sturmian
word ω, f (ω) is a Sturmian word.
Example 5.25. The morphisms identity 1 : a → a, b → b and swap
S : a → b, b → a are Sturmian. The morphisms

La : a → a, b → ab, Ra : a → a, b → ba

and

Lb : a → ba, b → b, Rb : a → ab, b → b

are Sturmian.

Definition 5.41 (Mignosi and Séébold, 1993). The Sturmian

monoid St is generated by (S, La , Ra ) (or by (S, Lb , Rb )).

Remark 5.4. The Sturmian monoid St is the submonoid of the

automorphisms group Aut(F2 ) of the free group F2 = {0, 1}. The
subgroup generated by (S, La , Ra ) is Aut(F2 ).
128 A Compendium of Musical Mathematics

By interpreting the letter a as the encoding of a tone and the letter

b as the representative of a semitone, the Sturmian morphisms by suc-
cessive actions generate the diatonic modes of Henricus Glareanus:
La La Lb (ab) = La La (bab) = La (abaab) = aabaaab = Ionian
La Ra Lb (ab) = La Ra (bab) = La (baaba) = abaaaba = Dorian
Ra Ra Lb (ab) = Ra Ra (bab) = Ra (baaba) = baaabaa = Phrygian
La La Rb (ab) = La La (abb) = La (aabab) = aaabaab = Lydian
La Ra Rb (ab) = La Ra (abb) = La (ababa) = aabaaba = Mixolydian
Ra Ra Rb (ab) = Ra Ra (abb) = Ra (ababa) = abaabaa = Aeolian
Definition 5.42. Two modes are said to be conjugated if the asso-
ciated words are conjugated.
Example 5.26. Ionian mode (u = aab, v = aaab) is conjugated
with Lydian mode. Dorian mode is conjugate of Mixolydian mode
and Phrygian mode is conjugate of Aeolian mode. In this model, the
Locrian baabaaa mode has no conjugate.
Proposition 5.14. Let f, g be two morphims. We have the following
statements:
(1) If f ◦ g is Sturmian and f Sturmian, then g is Sturmian.
(2) If f ◦ g is Sturmian and g Sturmian, then f is Sturmian.
The projection π : Aut(F2 ) → GL2 (Z) is defined by
0 1 1 1
π(S) = , π(La ) = π(Ra ) = R =
1 0 0 1
and
1 0
π(Lb ) = π(Rb ) = L =
1 0
The group SL2 (Z) is generated by L, R with relations given by
RL−1 R = L−1 RL (braid relations) and (RL−1 R)4 = 1 (torsion rela-
tion). These relations can be lifted to Aut(F2 ). We get the commu-
tation relations
La Ra = Ra La , Lb Rb = Rb Lb
 Combinatorics on Musical Words 129

the braid relations

La L−1 −1 −1
b La = Lb La Lb , La Rb−1 La = Rb−1 La Rb−1
Ra L−1 −1 −1
b Ra = Lb Ra Lb , Ra Rb−1 Ra = Rb−1 Ra Rb−1

and the torsion relations

(La L−1 4
b Ra ) = 1, (L−1 −1 4
b Ra Rb ) = 1

(Ra Rb−1 La )4 = 1, (Rb−1 La L−1 4

b ) =1

Definition 5.43. The braid group (or Artin group) Bn is defined by

generators σ1 , σ2 , . . . , σn−1 and by relations

σi σj = σj σi if |j − i| ≥ 2

and

σi σi+1 σi = σi+1 σi σi+1 for i = 1, 2, . . . , n − 1

Theorem 5.23. Let f be a group homomorphism f : B4 → Aut(F2 )

defined by f (σ1 ) = La , f (σ2 ) = L−1
b and f (σ3 ) = Ra . Then the
following sequence is exact
f
1 → Z(B4 ) → B4 −→ Aut(F2 ) → Z2 → 1

where Z(B4 ) is the center of the braid group B4 .

Theorem 5.24 (Kassel–Reutenauer). The following sequences
are exact:
f
1 / Z(B4 ) / B4 / Aut(F2 ) det / Z2 /1

 π π =
   
1 / 2Z(B3 ) / B3 / GL2 (Z) det / Z2 /1

The morphism π  : B4 → B3 is defined by π  (σ1 ) = π  (σ3 ) = R and

π  (σ2 ) = L−1 .
Definition 5.44. The special Sturmian monoid is defined by

SSt = {f ∈ St, det(f ) = +1} ⊂ Aut(F2 )

130 A Compendium of Musical Mathematics

Theorem 5.25. The special Sturmian monoid has for generators

La , Ra , Lb , Rb and for relations, for all n ≥ 1

La Lnb Ra = Ra Rbn La , Lb Lna Rb = Rb Ran Lb

Theorem 5.26. The monoid generated by σ1 , σ2−1 , σ3 , σ4−1 in B4 is

isomorphic to the special Sturmian monoid SSt.
Theorem 5.27. The word ω is a Christoffel word if and only if there
exists two Christoffel words ω = (u, v) such that the matrix

|u|a |v|a
∈ SL2 (Z)
|u|b |v|b

Furthermore, (u, v) is the standard factorization.

Example 5.27. The Christoffel word of slope 5/8

ω = C(5/8) = aabaababaabab
= (u, v) = (aabaabab, aabab)

and
|u|a |v|a 53
det = det = +1
|u|b |v|b 32

The words u = C(5/3) and v = C(2/3) form the standard factoriza-

tion of C(5/8).
Proposition 5.15. Let ω be a Christoffel word of slope s/t and (u, v)
its standard factorization. The words La (ω) = (u, uv) and Rb (ω) =
(uv, v) are Christoffel words.
Remark 5.5. The morphism La assigns the Christoffel word of slope
s/t into the Christoffel word of slope s/(s + t). And the morphism Rb
assigns the Christoffel word C(s/t) of slope s/t into the Christoffel
word C((s + t)/t). For instance, for ω = C(3/5) = aabaabab, we get

La (ω) = aaabaaabaab = C(3/8)

and

Rb (ω) = abaabababbabb = C(8/5)

 Combinatorics on Musical Words 131

Thomas Noll showed that in the 12-EDO, a word coded according

to tones (a) and semitones (b) can also be written in terms of ascend-
ing fifths (x) and descending fourths (y) on the alphabet {x, y}.
The correspondence between the word representing the mode on the
alphabet {a, b} and the word representing the same mode on the
alphabet {x, y} is called the Christoffel duality.

Example 5.28. Let N = 12, consider the word ω of ascending

fifths and descending fourths generated by the morphism x →
x + 7 mod 12:
y x y x y x y y x y x y
5 → 0 → 7 → 2 → 9 → 4 → 11 → 6 → 1 → 8 → 3 → 10 → 5

This word is a Christoffel word

ω = C(7/5) = (yxyxyxy)(yxyxy) = (u, v)

of standard factorization since

32
det = +1
43

The standard factorization of C(7/5) determines the white

{5, 0, 7, 2, 9, 4, 11} and black {1, 8, 3, 10} keys of the 12-EDO.

Definition 5.45. Let p, q be two nonnegative coprime integers and

N = p + q. Let p∗ , q ∗ ∈ {0, 1, . . . , p + q − 1} defined by pp∗ = 1 mod N
and qq ∗ = 1 mod N . The dual Christoffel word C(p/q) is the Christof-
fel word C(p∗ /q ∗ ) of the slope p∗ /q ∗ .

Example 5.29. Let p = 4, q = 7. The Christoffel word

ω = C(4/7) = xxyxxyxxyxy

has a dual given by p∗ = 3 and q ∗ = 8:

ω ∗ = C(3/8) = xxxyxxxyxxy

Proposition 5.16. Let ω be a proper Christoffel word C(p/q) of

standard factorization ω = uv. Then the lengths of u and v are related
to the dual Christoffel word of slope p∗ /q ∗ by |u| = p∗ and |v| = q ∗ .
132 A Compendium of Musical Mathematics

5.6 Plactic Congruences

In this section, we follow Jedrzejewski (2011). Musical scales are pat-

terns of five notes or more that move by steps. Since each kind
of step can be encoded by a letter of a totally ordered alphabet
A ⊂ ZN = {0, 1, . . . , N − 1}, the structure of a scale is represented
by a word ω over this alphabet such that the sum of each step is equal
to the total number N of notes in the N -tone equal temperament:

i=N
i∈ω

In the usual tempered system, the Dorian mode (D, E, F, G, A, B, C)

is encoded by the word 2122212, or babbbab, with a = 1, b = 2, a < b
and the major scale or ionian mode is represented by 2212221 or
bbabbba. If for some musical reasons a note is altered, the encod-
ing of the mode changes. Two modes are in the same class if
they are related by a chain of words in plactic relations. These
relations have been introduced by Knuth (1970) and the plactic
monoid of Schützenberger and Lascoux appears in (Lascoux and
Schützenberger, 1981; Schützenberger, 1997).
Definition 5.46. The plactic monoid over some totally order finite
alphabet A = {a, b, c, . . .} with a < b < c < · · · is the monoid
whose generators are the letters of the alphabet verifying for any
three arbitrary letters x, y, z in A the Knuth congruence relations

yzx ≡ yxz whenever x < y ≤ z
xzy ≡ zxy whenever x ≤ y < z
The plactic monoid on the alphabet A is the quotient of the monoid
of words A∗ by the Knuth relations Pl(A) = A∗ / ≡.
The plactic monoid of modes is the restriction of this set to the
modes.
Remark 5.6. If the alphabet has only two letters A = {a, b} with
a < b, the plactic relations reduce to
bab ≡ bba, aba ≡ baa
When a encodes a semitone and b a whole tone (a = 1 and b = 2),
two modes are in the same class if there exists a chain of modes that
 Combinatorics on Musical Words 133

differ by a pattern of 4 consecutive notes such that their steps are

bab ≡ bba or aba ≡ baa.
Example 5.30. For example, the Dorian mode (babbbab) and the
Mixolydian mode (bbabbab) are in the same class because the move
of the third note in the pattern (D, E, F, G) to (D, E, F , G) is
allowed by the plactic congruence. But there is no direct plactic
relations between the Carnatic Vachaspati mode (bbbabab) and the
Rishabhapriya mode (bbbaabb). Vaschaspati is linked to the Mixoly-
dian mode (bbabbab), which is linked to the Carnatic Charukesi mode
(bbababb), and this last mode is linked to Rishabhapriya.
If the alphabet has three letters A = {a, b, c} with a < b < c, the
plactic relations
bca ≡ bac, acb ≡ cab
described the switch of two steps (ac ∼ ca) if and only if this pattern
is preceded or followed by the step b (a < b < c). In other words, in
a four consecutive notes pattern, the second or the third note can be
raised or lowered if the first step is b. For the simplest values (a = 1,
b = 2, c = 3), the plactic relations mean that in a tritone 213, (F, G,
G , B), the third note can be raised (F, G, A , B) but the second note
cannot move (F, F , G , B). And if the last step is b = 2, in a tritone
312, (F, G , A, B), the second note can be lowered (F, F , A, B),
but the third note cannot move (F, G , A , B).

5.6.1 Robinson–Schensted–Knuth correspondence

In 1961, Schensted (1961) gave an algorithm to find the length of
the longest nondecreasing subwords of a given word w ∈ A∗ . His
method associates with each word ω a semistandard Young tableau
t = P (ω). The words producing a given Young tableau t form a
plactic class. Lascoux and Schützenberger (1981) have stated that the
plactic classes [ω] and [ω  ] of two words ω and ω  uniquely determine
the plactic class of their concatenation [ωω  ]. And this gives the set
of all plactic classes.
Schensted’s algorithm is as follows. In this section, a row is a
nondecreasing word. The elementary step of the algorithm consists
of the insertion of a letter x into a tableau t = u1 u2 · · · ul , where u1
is the first row and ul the last row. First, one tries to insert x into u1 .
134 A Compendium of Musical Mathematics

If u1 x is a row (a nondecreasing word), then P (u1 x) = u1 x, that is,

the letter x is added to the right of the first row. Otherwise, if u1 x is
not a row, we determine y the leftmost letter of u1 which is greater
that x, replace y by x in the row u1 , insert y into u2 with the same
rules (possibly recursively), and so on.
Example 5.31. The computation of the semistandard tableau
P(142531) is given by the following steps:
1 1 3
1 2 1 2 5 1 2 3 2 5
1 1 4 4 4 4 5 4

In 1961, Schensted stated that the length of the first row of P (ω)
is equal to the maximal length of nondecreasing subwords (in our
examples, the three subwords 145, 125, 123 have length 3) and the
length of the first column of P (ω) is equal to the maximal length
of the decreasing subwords. Some years later, Greene introduced the
equivalence relations on two words of A∗ if they have the same Schen-
sted tableau:
ω ∼ ω  ⇐⇒ P (ω) = P (ω  )
Moreover, in 1970, Knuth (1970) proved that the equivalence ∼
between two words coincides with the plactic congruence.
Another tool for studying the plactic monoid is the RSK
(Robinson–Schensted–Knuth) correspondence.
Definition 5.47. The RSK map is associate with each word ω ∈
A∗ over the linearly ordered alphabet A = {1, 2, . . .}, the pair
(P (ω), Q(ω)), where P (ω) is the Schensted semistandard Young
tableau and Q(ω) is a standard Young tableau.
Proposition 5.17. The RSK map ω → (P (ω), Q(ω)) is one-to-one.
A standard tableau is a semistandard tableau where each k ∈ A
appears exactly once. By definition, Q(ω) is the standard tableau
encoding the chain of shapes in the construction of P (ω). In the
previous example, Q(142531) is the tableau:
1 2 4
3 5
6
 Combinatorics on Musical Words 135

Let us denote m1 , m2 , ..mr the multiplicities of the letters a, b, c, . . .

in the word w. By labeling from left to right each occurrence of a with
label 1 to m1 , each occurrence of b with labels m1 + 1 to m1 + m2 ,
and so on, we define a new word called the standardization std(ω) of
the word w. For example,

std(142531) = 153642

Theorem 5.28. If two words are in Knuth relations ω ≡ ω  , then

they have the same standardization std(ω) = std(ω  ). Therefore, the
standardization commute with the Schensted tableau

P (std(ω)) = std(P (ω)), Q(ω) = Q(std(ω))

5.6.2 Plactic modal classes

As an application of the results of the previous section, we will classify
all modes with the plactic congruences on the alphabet A = {a, b}.
Definition 5.48. The evaluation of a word ω over the alphabet
A = {a1 , a2 , . . . , an } is the vector whose components are the number
of occurrences of each letter of A, namely

ev(ω) = (|ω|a1 , |ω|a2 , . . . , |ω|an )

where |ω|a1 indicates the number of occurrences of a1 in ω.

Definition 5.49. For any word ω = ω1 · · · ωn , a descent is a position
i, where ωi > ωi+1 . The major index maj(ω) is the sum of all descent
positions.
Example 5.32. For example, the word 23232 has descents in posi-
tion 2 and 4. Its major index is maj(ω) = 2 + 4 = 6.
Definition 5.50. The generating polynomial of a subset R ⊂ A∗ is
defined by

T (x) = xmaj(ω)
ω∈R

The generating function of the set R of the plactic classes related

to a fixed pair (a, b) decomposes into cyclotomic polynomials and
T (1) is the number of modes or the cardinality of the set R.
136 A Compendium of Musical Mathematics

The graph of a plactic class is defined as follows. The vertices of

this graph are the words representing the modes of the class. Two
vertices are connected by an edge if and only if the two words repre-
senting the vertices are related by Knuth relations.

Definition 5.51. A plactic modal class is defined to be a plactic class

of five or more modes which has no trivial linear graph, otherwise
the class is said linear plactic class.

If we consider all modes with less than 12 notes, a computation

shows that there are only 10 plactic modal classes, that is to say,
10 plactic classes with nontrivial graph. Table 5.1 gives the values
in the 12-tone chromatic system of the two constitutive intervals a
and b, the number of elements in the class and the expression of one
representative.

Remark 5.7. The previous plactic classes are unchanged if we

replace the 12-tone system by the N -tone equal temperament (except
for the values of a and b). In this tuning, new classes appear. For
instance, in the 12-EDO, there are 2 linear plactic classes of modes
of 11 notes: a class with 1 element a10 b and another class with
10 elements linked to ba10 . In the 13-EDO, the modes of 11 notes have
a new set composed of 3 classes: a trivial class with 1 element a9 b2 ,
a linear class of 10 elements with representative ba9 b and a modal
class of 44 elements with representative b2 a9 . Louvier’s heptatonic

Table 5.1. The ten first plactic modal classes.

Notes a b Representatives Elements

6 1 3 bbaaab 9
6 1 4 baaaba 9
7 1 2 bababbb 14
8 1 2 babababa 14
8 1 2 aabababb 20
8 1 2 abababab 28
8 1 3 aaaababa 20
9 1 2 aaaababab 27
9 1 2 aaabababa 48
10 1 2 aaaaaababa 35
 Combinatorics on Musical Words 137

mode 4334343 used in its Clavecin non tempéré belongs to a new

(nonlinear) class of 14 elements.
Remark 5.8. In the 24-EDO, Wyschnegradsky’s diatonicized chro-
matic scale {0, 1, 3, 5, 7, 9, 11, 12, 14, 16, 18, 20, 22} associated with
the mode 1222221222222 belongs to a linear class of 12 elements.
A new modal class of 65 elements appears in this tuning with repre-
sentative b11 a2 .
Remark 5.9. In the 36-EDO (third of tones universe), the diatonic
scale of 19 notes (ab8 ab9 , with a = 1, b = 2) belongs to a class of 18
elements. There is also a class of 152 elements. The modes with 12
notes lead to 3 classes of respectively 54, 154 and 275 elements.
Since plactic congruences defined modulations from one mode
to another, the plactic modal classes define the admissible domain
of modal modulations. These modulations, also called sometimes
metaboles, could be generalized to microtones. As mentioned above,
microtonal combinatorics, as well as plactic modes classification, give
new resources yet unexplored.

5.6.3 Pentatonic modes

To study the pentatonic modes, we distinguish between classes
according to their evaluation:
(1) If ev(ω) = (4, 1). Each word ω has four letters a and one
letter b. In the 12-tone equal temperament, the modes in semitones
correspond to either (a = 1, b = 8) or (a = 2, b = 4). Thus, there
are five modes. Since there are two shapes of semistandard tableaux,
the five modes decompose into two plactic classes. The first class has
only one element aaaab with insertion tableau P (ω) of the form
a a a a b
and the second class has four elements given by their Schensted
tableau P (ω)
a a a a
b
The plactic congruences define the plactic class of baaaa:
aaaba ∼ aabaa ∼ abaaa ∼ baaaa
138 A Compendium of Musical Mathematics

Their recording tableau Q(ω) are respectively

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

(2) If ev(ω) = (3, 2). Each word ω has three letters a and two
letters b. In the 12-tone equal temperament, the modes correspond
to a = 2, b = 3 semitones. Since there are three possible shapes,
there are three plactic classes. The first class has only one element
a a a b b

The second class has four elements given as follows. In this class, the
pentatonic minor mode (22323) is in relation with the Blues major
mode (23223):
aabba ∼ aabab ∼ abaab ∼ baaab
22332 22323 23223 32223
All elements of this class have the same Schensted tableau P (ω):
a a a b
b

Finally, the third class has five elements. In this class, the Egyptian
mode (23232) is linked to the pentatonic major mode (32232) which
is related by plactic congruence to the Blues minor mode (32322):
abbaa ∼ ababa ∼ baaba ∼ babaa ∼ bbaaa
23322 23232 32232 32322 33222
The Schensted tableau has the form
a a a
b b

Remark 5.10. If we switch to the N -tone equal temperament, the

classes remain unchanged. For instance, in the quarter tone universe
(N = 24), the evaluation ev(ω) = (4, 1) leads to 4 sets of 5 modes
obtained for (a, b) = (1, 20), (2, 16), (3, 12) and (4, 8). In the same
way, ev(ω) = (3, 2) yields 2 sets of 10 modes for (a, b) = (2, 9) and
(4, 6). But ev(ω) = (1, 4) produces one set of five modes for (a = 4,
b = 5) and ev(ω) = (2, 3) gives a set of ten modes with (a = 3,
b = 6).
 Combinatorics on Musical Words 139

Fig. 5.2. Plactic classification of hexatonic modes (class of babaab).

5.6.4 Hexatonic modes

The 462 hexatonic modes, distributed according to 80 hexatonic
scales, formed by two intervals (a = 1, b = 3 or a = 1, b = 4)
are grouped into linear classes with the exception of the babaab
mode class (C, D , E, F, G , A) which has the following structure (see
Fig. 5.2). The most famous of the hexatonic modes is the whole tone
scale (222222). In an 1860 letter to Ingeborg Stark, Liszt nicknamed
it the terrifying scale (Der traurige Mönch). It is used systematically
and almost exclusively in Debussy’s second piano prelude, Voiles,
composed in 1909, with some borrowings from the minor pentatonic
scale on its fourth degree pedal.

5.6.5 Heptatonic modes

Among the heptatonic modes, two classes have more than five ele-
ments. The first heptatonic class is the class of the Phrygian mode
(abbbabb). In the 12-EDO, this class is obtained for a = 1, b = 6 or
a = 1, b = 2. Its graph is linear (shown in the following) and it is not
a plactic modal class:

baabbbb
|
ababbbb
|
abbabbb − abbbabb − abbbbab − abbbbba
locrian phrygian natakapriya kikilapriya
140 A Compendium of Musical Mathematics

Fig. 5.3. Plactic classification of heptatonic modes (class of bbabbba).

The second class is the plactic class of the major mode (bbabbba
with a = 1, b = 2). These modes appear in John Foulds’s classifica-
tion under the Vth class, and three of them are used in the Essays
in the modes for piano (1920–1927). The plactic heptatonic class has
14 elements and its graph is as follows (see Fig. 5.3).

5.6.6 Octatonic modes

The 330 octatonic modes are divided into 43 scales. Perhaps the
most famous octatonic mode is Messiaen’s mode 2 which is known
by different names and which Scriabin made extensive use in his
Seventh Piano Sonata (1912). The plactic class of mode 2 consists
of 14 elements issued from 21212121 and 28 elements issued from
12121212 (see Fig. 5.4).
Two other important modal classes have a similar structure. The
class of mode 21112221 consists of 20 elements and the class of mode
13111311 consists of 20 elements.
 Combinatorics on Musical Words 141

Fig. 5.4. Plactic classification of octatonic modes (class of 12121212).

5.7 Rational Associahedra

Rational associahedra have been introduced by Drew Armstrong

(Armstrong et al., 2013). A rational associahedron is a simplicial
complex based on a generalization of Catalan numbers called ratio-
nal Catalan numbers. In this section, we follow Jedrzejewski (2017a).
Definition 5.52. For all rational number x ∈ Q\[−1, 0] written
r
uniquely as x = where r and s are coprime positive integers
s−r
r = s, the rational Catalan number is defined by the formula
(r + s − 1)!
Cat(x) = Cat(r, s) =
r!s!
For a positive integer x = n, we recover the classical Catalan
numbers:
Cat(n) = Cat(n, n + 1)
If we define a rational Catalan design as a combinatorial design whose
number of blocks b is a rational Catalan number Cat(x) for some x
142 A Compendium of Musical Mathematics

positive rational number, we see that all combinatorial designs can

be considered as a rational Catalan design. The reason is that we
can always find x such that b = Cat(x). It is enough to take x =
2/(2b − 1), (r = 2, s = 2b − 1). That is to say, with this definition, we
can always reach a combinatorial design of b = (x+1)/2x blocks. The
difference between the definition of a rational design with a Catalan
number and this new definition with a rational Catalan number is
that with the latter we can have several values of x for a given b.
For instance, the rational Catalan number of 14 blocks is obtained
for the integer x = 4, with r = 4, s = 5 (the classical case), and also
for x = 2/25, with r = 2 and s = 27. Note that for a given b, one
value of x is not necessarily an integer, for instance, Cat(x) = 12, x
is x = 2/21 and x = 3/4. In some cases, we get only one solution:
Cat(x) = 16, x = 2/29.
Furthermore, we have the following:
Definition 5.53. The derived Catalan Numbers are defined by
1 x
Cat (x) = Cat = Cat
x−1 1−x
1 x
An important property derives from that 1/x−1 = 1−x and hence

Cat (x) = Cat (1/x)

Another important property is that rational Catalan numbers are in

relation with Dyck paths.
Definition 5.54. A (r, s)-Dyck path is a staircase walk on the integer
lattice Z2 from (0, 0) to (r, s) that lies strictly above (but may touch)
the diagonal y = rx/s.
It has been shown that
Proposition 5.18. The number of (r, s)-Dyck paths is Cat(r, s).
We recover a classical result that the number of (classical)
Dyck paths (staircase from (0, 0) to (n, n)) is the Catalan number
Cat(n, n + 1) = Cat(n).
Associated with these rational Catalan numbers exists a geometric
structure that D. Armstrong called rational associahedron Ass(r, s)
which has properties similar to the classical case of associahedron.
 Combinatorics on Musical Words 143

In particular, there exist bijections between the set of homogeneous

(r, s)-noncrossing partitions and the set of (r, s)-Dyck paths (see
Armstrong et al., 2013 for details).

Definition 5.55. A rational associahedron Ass(r, s) is the simplicial

complex consisting of all noncrossing dissections of the regular s-gone
Ps+1 , s ≥ 2.

Definition 5.56. The facets are collections F (D) of admissible diag-

onals of Ps+1 , where D is a (r, s)-Dyck path.

Proposition 5.19. The number of facets in Ass(r, s) is Cat(r, s).

Theorem 5.29. A diagonal of Ps+1 which separates i vertices from

s − i − 1 vertices appears as a vertex of Ass(r, s) if and only if i ∈
S(r, s)

is
S(r, s) = ,1 ≤ i < r
r

where ξ is floor(ξ), the greatest integer ≤ ξ.

Example 5.33. For instance, if x = 3/2, r = 3 and s = 5, the ratio-

nal associahedron Ass(3, 5) has 6 vertices, since S(3, 5) = {1, 3}.
Ass(3, 5) has 7 facets as there are Cat(3, 5) = 7 Dyck paths. The
associahedron is drawn in Fig. 5.5. Each vertex represents an admis-
sible dissection of P6 . On this graph, a facet links two vertices. Each
facet corresponds to a (3, 5)-Dyck path, a staircase from (0, 0) to
(3, 5). Symbol x represents up step and y represents right step. The
simplicial complex Ass(r, s) has dimension r − 2.

The h-vector is the integer sequence (h−1 , h0 , . . . , hr−2 ) where

hi−2 is the number of (r, s)-Dyck paths with i nontrivial vertices
runs. A vertical run in a Dyck path is the maximal sequence of (0,
1) steps. hi−2 is also the number of Dyck paths with (i − 1) “val-
leys”, according to the shape of Dyck path. It is also the number of
monomial y m in the Dyck word minus 1. It is the Narayana number
given by

1 r s−1
hi−2 = Nar(r, s, i) =
r i i−1
144 A Compendium of Musical Mathematics

For the noncrossing partitions point of view, the number of (r, s)-
Dyck paths with αj vertices runs of length j is the Kreweras number:

(s − 1)!
Krew(r, s, α) =
α0 !α1 ! · · · αr !
The f -vector is the integer sequence (f−1 , f0 , . . . , fr−2 ) of Ass(r, s)
with f−1 = 1, and fi is the number of i-dimensional faces 0 ≤ i ≤
r − 2, also given by the Kirkman numbers:
1 r s+i−1
fi−2 = Kir(r, s, i) =
r i i−1
These numbers verify the relation


r−2 
r−2
fi (t − 1)r−2−i = hi tr−2−i
i=−1 i=−1

and the reduced Euler characteristic defined from the f -vector by

the alternate sum of its components is equal to the derived Catalan
number:

r−2
χ= (−1)i fi = (−1)r Cat (r, s)
i=−1

Example 5.34. For instance, for the Ass(3, 5) of Fig. 5.5, h-vector
= (1, 4, 2), f -vector = (1, 6, 7) and the reduced Euler characteristic
χ = −2 which is equal to (−1)3 Cat (3/2) = −Cat(2) = −2.
Dyck paths are related by plactic relations. Dyck path of Ass(3, 5)
is a class of this monoid with three letters x and five letters y, namely

x3 y 5 ≡ x2 yxy 4 ≡ xyx2 y 4 ≡ xyxyxy 3 ≡ x2 y 2 xy 3

≡ x2 y 3 xy 2 ≡ xyxy 2 xy 2

The blocks of the (7, 3, 1) design (see Chapter 8) are represented in

Fig. 5.5. One passes from one block to another by adding 2 to each
element of the block.
 Combinatorics on Musical Words 145

Fig. 5.5. Rational associahedron Ass(3, 5).

Example 5.35. The design (9, 3, 1) has 12 blocks and 4 parallel

classes:

000011122236
134534534547
268787676858

Since 12 = Cat(3, 7), the rational associahedron Ass(3, 7) with

8 vertices and 12 facets could be used to represent the design (9, 3, 1).
On the dissection plane P8 , each vertex separates 2 vertices from 4
vertices. Dyck paths lead to 12 facets. Each facet is associated with
a block of the design. These blocks are depicted in Fig. 5.6. The fig-
ure is a Möbius strip (glue the ribbon with respect to the arrows).
146 A Compendium of Musical Mathematics

Fig. 5.6. Möbius strip of design (9, 3, 1).

At each vertex, there are three blocks which form a parallel class
(a partition of Z9 ), except when the ribbon wraps.

5.8 Exercises

Exercise 5.1
Write down the Euclidean string E(4, 7). (It is the Bulgarian rhythm
called Ruchentiza).

Exercise 5.2
Is the Arab rhythm Agsag-Samai 12222 a Euclidean string?

Exercise 5.3
For integer n ≥ 2, write down the Euclidean string E(n, 2n + 1),
which is the hierarchy of some Aksak rhythms.

Exercise 5.4
Give two different proofs that Euclidean rhythm E(4, 9) is E-deep.
Proof that the Aksak rhythms hierarchy is E-deep.

Exercise 5.5
Fume-fume is an African bell pattern of 12 pulses {0, 2, 4, 7, 9}12 .
Show that fume-fume is well-formed, maximally even and deep.

Exercise 5.6
Show that the Bulgarian rhythm E(7, 18) = 3232323 has the rhyth-
mic oddity property (1) by using the definition (2) by using Theorem
5.13 and (3) by using Theorem 5.16.
 Combinatorics on Musical Words 147

Exercise 5.7
Using Theorem 5.16, show that the word ω = 332232322 has the
rhythmic oddity property.
Exercise 5.8
Let us consider the Christoffel word ω = C(8/3) on the alphabet
{a, b}. By drawing the Dick path, write down the standard factor-
ization of ω = (ω1 , ω2 ). Verify that

|ω1 |a |ω2 |a
∈ SL2 (Z)
|ω1 |b |ω2 |b

Write down its palindromic factorization. By interpreting ω has a

generated scale in Z11 (11 = 8 + 3), show that ω1 are the black keys
and ω2 the white keys of a tuning system with 11 degrees and a
diatonic scale defined in Chapter 3.

References

Armstrong, D., Rhoades, B. and Williams, N. (2013). Rational associahedra

and noncrossing partitions, The Electronic Journal of Combinatorics 20(3),
1–27.
Arom, S. (1991). African Polyphony and Polyrhythm, Cambridge University
Press, Cambridge.
Berstel, J., Lauve, A., Reutenauer, C. and Saliola, F. (2008). Combinatorics on
Words: Christoffel Words and Repetition in Words, CRC Monograph Series,
Vol. 27, American Mathematical Society, Providence.
Besada, J.L. (2017). Metamodels in Compositional Practices, The Case of Alberto
Posadas’s Liturgia Fractal, Ircam/Delatour, Sampzon.
Bouchet, A. (2010). Imparité rythmique, Culturemath, ENS, Paris.
Brăiloiu, C. (1952). Le rythme Aksak, Revue de musicologie 33, 71–108.
Chemillier, M. and Truchet, C. (2003). Computation of words satisfying the
“rhythmic oddity property”, Information Processing Letters 86(5),
255–261.
Chemillier, M. (2004). Grammaires, automates et musiques. In: Briot, J.P. and
Pachet, F. (eds.) Informatique musicale, Hermès, Paris, 195–230.
Clampitt, D. and Noll, T. (2011). Modes, the height-width duality, and Hand-
schin’s tone character, Music Theory Online 17(1), 1–43.
Clampitt, D. and Noll, T. (2018). Naming and ordering the modes, in light of
combinatorics of modes, Journal of Mathematics of Music 12(3), 134–153.
Ellis, J., Ruskey, F., Sawada, J. and Simpson, J. (2003). Euclidean strings,
Theoretical Computer Science 301, 321–340.
148 A Compendium of Musical Mathematics

Erdős, P. (1989). Distances with specified multiplicities, American Mathematical

Monthly 96, 447.
Gamer, C. (1967). Deep scales and difference sets in equal-tempered systems.
In: Proceedings of the 2nd Annual Conference of the American Society of
University Composers, 113–122.
Jedrzejewski, F. (2011). Plactic classification of modes. In: Agon, C. et al. (eds.)
Mathematics and Computation in Music, Third International Conference,
MCM 2011, Springer, 350–354.
Jedrzejewski, F. (2017a). Associahedra, combinatorial block designs and related
structures. In: Peck, R. and Montel, M. (eds.) Mathematical Music Theory,
World Scientific, Singapore, 47–64.
Jedrzejewski, F. (2017b). New investigations on rhythmic oddity. In Agustı́n-
Aquino, O.A., Lluis-Puebla, E. and Montiel, M. (eds.) Mathematics and
Computation in Music, 6th International Conference, MCM 2017, Proceed-
ings, Mexico City, Mexico, June 26–29, 2017, Springer, 227–237.
Justin, J. and Pirillo, G. (1997). Decimations and sturmian words, Informatique
théorique et applications 31(3), 271–290.
Kassel, C. and Reutenauer, C. (2007). Sturmian morphisms, the braid group B4 ,
Christoffel words and bases of F2 , Annali di Matematica Pura ed Applicata
186(2), 317–339.
Knuth, D.E. (1970). Permutations, matrices, and generalized Young tableaux,
Pacific Journal of Mathematics 34(3), 709–727.
Lascoux, A. and Schützenberger, M.P. (1981). Le monoı̈de plaxique. In: Non-
commutative structures in Algebra and Combinatorics, Quaderni della
Ricerca Scientifica del CNR 109, Roma.
Lascoux, A., Leclerc, B. and Thibon, J.-Y. (2002). The plactic monoid. In:
Lothaire, M. (ed.) Algebraic Combinatorics on Words, Cambridge Univer-
sity Press, Cambridge, 164–196.
Lerdahl, F. and Jackendorff, R. (1996). A Generative Theory of Music, MIT Press,
Cambridge.
Lothaire, M. (2002). Algebraic Combinatorics on Words, Cambridge University
Press, Cambridge.
Manoury, P. (2012). Les grammaires musicales génératives. www.philipppe.
manoury.com.
Noll, T. (2009). Ionian theorem, Journal of Mathematics and Music 3(3),
137–151.
Reutenauer, C. (2019). From Christoffel Word to Markoff Numbers, Oxford Uni-
versity Press, Oxford.
Schensted, C. (1961). Longest increasing and decreasing sequences, Canadian
Journal of Mathematics 13, 179–191.
Schützenberger, M.P. (1997). Pour le monoı̈de plaxique, Mathématiques Informa-
tique et Sciences Humaines 140, 5–10.
Steedman, M. (1984). A Generative grammr for jazz chord sequences, Music Per-
ception 2(1), 52–77.
 Combinatorics on Musical Words 149

Taslakian, P. (2008). Musical Rhythms in the Euclidean Plane, McGill University,

Montréal.
Toussaint, G.T. (2013). The Geometry of Musical Rhythm, Chapman & Hall,
CRC Press, Boca Raton.
Winograd, T. (1966). An Analysis of the Properties of Deep Scales in t-Tone
System, Unpublished, Colorado College, May 17, 1966.
This page intentionally left blank
 Chapter 6

Rhythmic Canons

The first canons appeared in the 12th century. The french chace
and the italian caccia are primitive forms of canons. Despite the
fact that canon is an old procedure, it has been used at any time,
from Renaissance composers (G. Ockeghem, M. Forestier, etc.) to
21st century composers. Famous canons have been composed by
Jean-Sébastien Bach (Musical Offering, Goldberg Variations, Canon-
ical Variations Vom Himmel hoch BWV 769 ), Gabriel Fauré
(Romances sans paroles, Thèmes et variations, Prélude 6 ), Arnold
Schoenberg (Opus 25, 28, 40, 41), Alban Berg (Kammerkonzert),
Anton Webern (Opus 2, 16, 28, 30, 31), Igor Stravinsky (Octuor,
Sextuor, In memoriam Dylan Thomas), Paul Hindemith (Fuga
undecima in B in Ludus Tonalis) and many others. For example,
Olivier Messiaen uses a rhythmic canon in the seventh song of Harawi
called Adieu. The rhythms are counted relatively to the double qua-
ver (e.g. 3 means three double quavers). The upper voice juxtaposes
three nonretrogradable rhythms: 35853, 43534 and 2235322. The
same rhythms appear in the two other voices shifted in time. From
a mathematical point of view, canons are linked to the decomposi-
tion of finite Abelian groups. Some results have been set by Amiot
(2005, 2011). Enumeration of rhythmic canons have been studied by
Fripertinger (2001).

151
152 A Compendium of Musical Mathematics

6.1 Tilings

Let’s start with the most general situation. Let G be a locally com-
pact Abelian group and G  its dual group. Let Ω be a subset of G of
finite nonzero Haar measure.
Definition 6.1. The set A is a tile if there exists a set B in G, called
the tiling set, such that
G=A⊕B
In other words, if every element g of G has unique decomposition,
g = a + b with a ∈ A and b ∈ B.
The pair (A, B) is called a tiling pair, or a factorization of the
group G. B is sometimes called a complementary tile to A. Usual
cases are G = Rd , R2 , R, Z and ZN = Z/N Z. In the first approach,
only the cyclical group ZN interests the musicians.
Example 6.1. The tile A = {0, 2} tiles the set Z4 = {0, 1, 2, 3} with
translations B = {0, 1}, i.e. with the sets A and A + 1 = {1, 3}.
This tiling corresponds to the decomposition of Z4 in the direct sum
Z4 = A ⊕ B = {0, 2} ⊕ {0, 1} = 2 · I2 ⊕ I2
where IN is the least residue system modulo N:
IN = {0, 1, . . . , N − 1}
In the musical domain, a factorization (A, B) is called a canon or
a tiling canon. The set A represents the set of events of the ground
voice (dux ) and the set B represents the time attacks: A + b with
b ∈ B are the different voices (comes) of the canon. The exchange
between the inner and the outer voices leads to the same results:
if (A, B) is a canon, then (B, A) is a canon and conversely.
Remark 6.1. If (A, B) is a canon of ZN , then |A| × |B| = N . In par-
ticular, |A| and |B| are divisors of N and thus the product of disjoint
sets of prime factors of N .
 Rhythmic Canons 153

6.1.1 Mask polynomials

Definition 6.2. The mask polynomial of a multiset A of nonnegative
integers is

A(x) = xa
a∈A

When working with polynomials, the direct sum of sets is inter-

preted as a product of polynomials and the union of sets as a sum of
the corresponding polynomials.
Example 6.2. In the previous example, A = {0, 2} is associated
with A(x) = x2 + x0 = x2 + 1 and the set B = {0, 1} corresponds to
the polynomial B(x) = x + 1. The product

A(x)B(x) = (x2 + 1)(x + 1)

= x3 + x2 + x + 1

is the characteristic polynomial of Z4 = A ⊕ B.

Proposition 6.1. The mask polynomials have the following
properties:
(1) The cardinal of the set A is the value of the mask polynomial at
point 1, namely |A| = A(1).
(2) Let Pm (x) be the mask polynomial of A in Zm . Then the mask
polynomial of the set kA in Zm is Pm (xk ).
(3) If X = A ⊕ B, then the mask polynomial of X is X(x) =
A(x)B(x) and the cardinal of X is the product

|X| = |A| × |B| = A(1)B(1)

(4) If X = A ∪ B, then the mask polynomial of X is the sum X(x) =

A(x) + B(x).
Remark 6.2. The canons are related to the underlying group (which
defines the sum). For instance, the tile 10110100 associated with
154 A Compendium of Musical Mathematics

A = {0, 2, 3, 5} tiles the integers Z but not the nonnegative integers

N (there is always a hole in position 1). The tile A tiles the group
Z8 = {0, 1, 2, . . . , 7} equipped with the sum modulo 8 but not the
same set I8 = {0, 1, 2, . . . , 7} equipped with the usual sum. The first
example is a tiling of the circle, whereas the second is a tiling of the
line.
Definition 6.3. Two canons E = A ⊕ B and E  = A ⊕ B  of ZN are
equivalent if |A| = |A |, |B| = |B  | and if there exists a translation
Tj and a permutation σ in the symmetric group S|T | such that
Tj (A + bi ) = A + bσ(i)
for i = 1, .. |T |. In other words, E and E  are equivalent, if the voices
of the canon E taken at any time are a permutation of the voices of
the canon E  .
Example 6.3. The canon E = {0, 1} ⊕ {0, 2} is equivalent to the
canon E  = {0, 3}⊕{0, 2} because T3 (A) = A and T3 (A+2) = A +2.

6.1.2 Cyclotomic polynomials

Recall some properties of the cyclotomic polynomials.
Definition 6.4. The n-th cyclotomic polynomial is defined by
  k
Φn (x) = (x − ζ) = (x − e2iπ n )
ζ 1≤k≤n
gcd(k,n)=1

where ζ is a primitive n-th root of unity in C.

Example 6.4. The first cyclic polynomials are Φ1 (x) = x − 1,
Φ2 (x) = x + 1, Φ3 (x) = x2 + x + 1, Φ4 (x) = x2 + 1, etc.
The cyclotomic polynomial Φn (x) is a monic polynomial with inte-
ger coefficients of degree ϕ(n), where ϕ is the Euler totient function.
Proposition 6.2. Primitive roots and cyclotomic polynomials are
linked by the relation

xn − 1 = Φd (x)
d|n
 Rhythmic Canons 155

Proposition 6.3. We have


1 + x + · · · + xn−1 = Φd (x)
d>1,d|n


d|n Φd (x) = x − 1 =
Proof. The proposition follows from n

(x − 1)(1 + x + · · · + xn−1 ) and the fact that Φ1 (x) = x − 1. 

Proposition 6.4. If n = p is prime, then

Φp (x) = 1 + x + x2 + · · · + xp−1

and
α
Φpα+1 (x) = Φp (xp )

Proof. As d|p Φd (x) = xp − 1 = Φ1 (x)Φp (x), so

xp − 1 xp − 1
Φp (x) = = = 1 + x + x2 + · · · + xp−1
Φ1 (x) x−1

To prove the second assertion, we have


α+1  α+1 α+1
Φ1 (x) Φpk (x) = Φs (x) = xp − 1 = (x − 1)(xp + · · · + 1)
k+1 s|pα+1

Simplifying by Φ1 (x) = (x − 1),

α+1 −1
1 + x + x2 + · · · + xp = Φp (x)Φp2 (x) · · · Φpα (x)Φpα+1 (x)

It follows

α+1
1 + x + x2 + · · · + xp −1
Φpα+1 (x) =
Φp (x)Φp2 (x) · · · Φpα (x)
 α α α  α 
1 + xp + x2p + · · · + x(p−1)p 1 + x + x2 + · · · + xp −1
=
1 + x + x2 + · · · + xpα −1
α α α α
= 1 + xp + x2p + · · · + x(p−1)p = Φp (xp )

156 A Compendium of Musical Mathematics

Proposition 6.5. Let p be prime. Then

⎧
(1) ⎪
⎨ 0 if s = 1
Φs (1) = q if s is a power of a prime q
⎪
⎩
1 otherwise

(2) Φps (x) if p is a factor of s

Φs (xp ) =
Φs (x)Φps (x) otherwise

(3) If s and t are relatively prime, then


Φs (xt ) = Φrs (x)
r|t

Proof.

(1) If s = 1, Φs (x) = x−1 and Φ1 (1) = 0. If s is a power of a prime q,

since Φq (x) = 1+x+x2 +· · ·+xq−1 , it implies that Φq (1) = 1 and
α−1
from Φqα (x) = Φq (xq ), we get Φqα (1) = Φq (1) = 1. Otherwise,
s have a prime decomposition s = pr11 · · · prnn . The product breaks
down into a product where d is a prime power and a product
where d has at least two primes. But by induction of the length
of the prime decomposition n, the last product is 1, so it remains
the product over prime powers
 
s= Φd (1) = Φs (1) Φd (1)
d>1, d|s d=prime power, d|s


n 
rj

n 
rj
= Φs (1) Φpl (1) = Φs (1) plj
j
j=1 l=1 j=1 l=1


n
r
= Φs (1) pj j = Φs (1)s
j=1

thus s = 1.
 Rhythmic Canons 157

(2) The roots of Φs (xp ) are e2iπk/ps for k relatively prime to s. If p is

a factor of s, then k is relatively prime to ps and e2iπk/ps are the
roots of Φps (x). Thus, Φs (xp ) = Φps (x). If p is not a factor of s,
either k is prime to ps or k is a multiple of p. If k is a multiple
of p, k = pr and gcd(r, p) = 1. It follows that e2iπk/s are the
roots of Φs (x) and since r is coprime to s, e2iπpr/s are the roots
of Φs (xp ). Hence, Φs (xp ) = Φs (x) and Φs (xp ) has the same roots
as Φs (x)Φps (x).
(3) Repeated application of (2) yields (3). Let t = p1 · · · pm+1 , where
some primes can be repeated, then
Φs (xt ) = Φs (x(p1 ···pm )pm+1 ) = Φs (xp1 ···pm )Φpm+1 s (xp1 ···pm )
 
= Φrs (x) Φrpm+1 s (x)
r|p1 ···pm r|p1 ···pm
 
= Φrs (x) = Φrs (x)
r|p1 ···pm+1 r|t 

6.1.3 Basic properties of tilings

Proposition 6.6. Let N be a positive integer and A,B are multisets
of nonnegative integers. The following statements are equivalent:
(1) A ⊕ (B ⊕ N Z) = Z.
(2) A ⊕ B = ZN , i.e. A ⊕ B is a complete set of representatives
of ZN .
(3) A(x)B(x) ≡ 1 + x + · · · + xN −1 mod (xN − 1).
(4) A(1)B(1) = N and for every factor s > 1 of N , the cyclotomic
polynomial Φs (x) divides A(x) or B(x).
Proof. (1) ⇔ (2) is trivial since ZN ⊕ N Z = Z.
(2) ⇒ (3) is trivial since
A(x)B(x) = 1 + x + · · · + xN −1
Therefore, A(x)B(x) ≡ 1 + x + · · · + xN −1 mod (xN − 1).
(3) ⇒ (2). The equality
 
A(x)B(x) = xa+b ≡ xa+b mod N mod (xN − 1)
a∈A, b∈B a∈A, b∈B

implies that A ⊕ B = ZN .
158 A Compendium of Musical Mathematics

(3) ⇒ (4). If (3), then there exists a polynomial Q(x) such that

A(x)B(x) = (xN − 1)Q(x) + (1 + x + · · · + xN −1 )

Thus, for x = 1, we get A(1)B(1) = 0 + N . Moreover, since Φs (x) is

irreductible, it must divide either A(x) or B(x).
(4) ⇒ (3). Consider the Euclidean decomposition

A(x)B(x) = Q(x)(xN − 1) + R(x)

for some integer polynomials Q(x), R(x) with deg(R) < N . Since
the product A(x)B(x) is by hypothesis divisible by the product of
all Φs (x), it is divisible by 1 + x + · · · + xN −1 . It follows that R(x) =
C(1 + x + · · · + xN −1 ), where C is a constant. Taking x = 1 leads to
A(1)B(1) = R(1) = N = CN . Therefore, C = 1, and

A(x)B(x) = Q(x)(xN − 1) + (1 + x + · · · + xN −1 )
≡ (1 + x + · · · + xN −1 ) mod (xN − 1)


Definition 6.5. Let A be a set of ZN . The difference set is the set

of all difference between elements

A − A = {a − a ; a, a ∈ A}

Proposition 6.7. Let N be a positive integer and A,B are nonempty

subsets of ZN . The following statements are equivalent:
(1) (A, B) is a tiling pair of ZN , i.e. A ⊕ B = ZN .
(2) ZN = A + B and N = |A| |B|.
(3) |A| |B| = N and (A − A) ∩ (B − B) = {0}.
(4) The sets A + b, b ∈ B form a partition of ZN .
Proof. (1) ⇔ (2) is trivial since it is the definition of the factori-
sation of a finite Abelian group.
(2) ⇒ (3). Suppose that (A − A) ∩ (B − B) = {g}. As g belongs
to A − A and to B − B, there exists a, a ∈ A and b, b ∈ B such that
g = a − a = b − b , that is, a + b = a + b. Now a = a and b = b ,
otherwise if a = a , the elements of the sum A+ B are not all distinct
and N = |A + B| < |A| |B|. Therefore, g = 0.
 Rhythmic Canons 159

(3) ⇒ (2). The map A × B → ZN , (a, b) → a + b is one-to-one

from the hypothesis. But it is also onto because N is the product of
the cardinality of A and B.
(3) ⇒ (4). From (A − A) ∩ (B − B) = {0}, it follows that the
sets A + b, b ∈ B are pairwise disjoint. Otherwise, choose b = b ,
b, b ∈ B such that (A + b) ∩ (A + b ) = ∅. Suppose there is an
element g ∈ (A + b) ∩ (A + b ). As g ∈ A + b, there is an a ∈ A such
that g = a + b. As g ∈ A + b , there is an a ∈ A such that g = a + b .
Thus, a + b = a + b , and so a − a = b − b . But this implies that
a − a = 0, b − b = 0, which contradicts that b and b are different.
Moreover, from ZN = A + B, it follows that A + b, b ∈ B form a
partition of ZN .
(4) ⇒ (1).
 
|A + B| = (A + b) = |A + b| = |A| |B|
b∈B b∈B

We define the Fourier transform of the set A ⊂ ZN , as the Fourier
transform of its characteristic function


A(n) = F(1A )(n) = e−2iπa/N
a∈A


and let Z(A) the set of the zeroes of A

Z(A) = {k ∈ ZN , A(k) = 0}

Converting the previous proposition in the language of Fourier trans-

forms, we get the following:
Proposition 6.8. Let N be a positive integer and A, B nonempty
subsets of ZN . A ⊕ B = ZN if and only if the zeroes of the Fourier
transforms of A and B cover ZN \{0}

Z(A) ∪ Z(B) = ZN \{0}

and N = |A| |B|.

Proposition 6.9. If the set A tiles ZN with B, then any motif C
homometric with A tiles ZN .
160 A Compendium of Musical Mathematics

6.1.4 Perfect rhythmic tilings

Perfect Rhythmic Tilings are particular tilings in which the tile is
augmented by affine transformations. They have been extensively
studied by Jean-Paul Davalan (2011). Composer Tom Johnson used
them in some pieces such as Tilework for piano (2003) or Plucking
(2003–2016), and Dean Rosenthal used one tiling of index 4 and order
15 in his Tiling for String Instrument (2011).
Let Ma,b be the affine transformation of ZN
Ma,b : x → ax + b mod N
Definition 6.6. Let n, k be two nonnegative integers. A perfect
rhythmic tiling of length N = nk and of index k is a partition of
ZN in n sets of k elements Ai,j = Mi,j ({0, 1, . . . , k − 1}) with strictly
positive different integers, whose union is the set ZN of the nk first
integers:

ZN = Ai,j
(i,j)∈S

such that all coefficients i are distinct. The multiplicative coefficients

i are called the augmentations and S the set of affine coefficients of
the tiling.
Example 6.5. For n = 5 and k = 3, consider the set of affine
coefficients
S = {(1, 2), (2, 4), (4, 5), (5, 1), (7, 0)}
and the tiling of

Z15 = Mi,j ({0, 1, 2})
(i,j)∈S

given by the following table. The set Z15 is tiled by five augmentations
of A = {0, 1, 2}:

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
A+2 0 0 1 1 1 0 0 0 0 0 0 0 0 0 0
2A + 4 0 0 0 0 0 0 0 0 1 0 1 0 1 0 0
4A + 5 0 0 0 0 0 1 0 0 0 1 0 0 0 1 0
5A + 1 0 1 0 0 0 0 1 0 0 0 0 1 0 0 0
7A 1 0 0 0 0 0 0 1 0 0 0 0 0 0 1
 Rhythmic Canons 161

This tiling, used in Tilework for piano, can also be written in a

compact form, keeping only the multiplicative coefficient of the affine
transformation:

751114572425247

6.2 Tijdeman’s Theorem

N. de Bruijn and G. Hajós stated the following theorem.

Theorem 6.1. If A is a finite set and A ⊕ C = Z, then there exists
a finite set B such that C = B ⊕ N Z, with N = |A| |B|.
Lemma 6.1. Let A and B be finite sets of nonnegative integers and
let N = A(1)B(1). If

A(x)B(x) ≡ 1 + x + · · · + xN −1 mod (xN − 1)

and p be prime which is not a factor of A(1), then

A(xp )B(x) ≡ 1 + x + · · · + xN −1 mod (xN − 1)

Theorem 6.2 (Tijdeman). If A is a finite set, 0 ∈ A ∩ C and

A ⊕ C = Z. If r and |A| are relatively prime, then rA ⊕ C = Z.
Proof. Let p1 p2 · · · pn a prime decomposition of r. By the previous
theorem, there exists a finite set B such that C = B ⊕ N Z, and
A ⊕ B = ZN . Then, A(x)B(x) ≡ 1 + x + · · · + xN −1 mod (xN − 1)
and by the previous lemma,

A(xp1 )B(x) ≡ 1 + x + · · · + xN −1 mod (xN − 1)

Applying this lemma repeatedly, we get

A(xp1 ...pn )B(x) ≡ 1 + x + · · · + xN −1 mod (xN − 1)

Therefore,

A(xr )B(x) ≡ 1 + x + · · · + xN −1 mod (xN − 1)

and thus rA ⊕ B = ZN since A(xr ) = rA(x). It follows rA ⊕ (B ⊕

N Z) = Z and thus rA ⊕ C = Z. 
162 A Compendium of Musical Mathematics

Corollary 6.1. Let A ⊕ B = ZN for some N. If k is relatively prime

to N , then A ⊕ kB = ZN , where kB = {kb, b ∈ B} is the dilation of
B by k.

Tijdeman’s theorem has many consequences. For instance, we can

deduce the following proposition.

Proposition 6.10. If a finite set A tiles the integers, then there is a

tiling by A whose period is a product of powers of the prime factors
of A.

Proof. Consider the tiling A ⊕ C = Z of period n where A is a

finite set and let r > 1 a factor of period n relatively prime to |A|. By
applying Tijdeman’s theorem, rA⊕ C = Z. Therefore, rA⊕ C0 = rZ,
where C0 is the set C0 = {c ∈ C, c ≡ 0 mod r}. Hence, A⊕C0 /r = Z
is a tiling of period n/r. 

Some remarkable properties of finite cyclic groups of prime power

order have been stated (Szabo and Sands, 2009).

Theorem 6.3. Let G be a finite cyclic group of prime power order

N such as G = A ⊕ B. If FN (x) = 1 + x + · · · + xN −1 divides A(x),
then A is periodic.

Corollary 6.2. In any factorization of a cyclic group of prime power

order, one factor is always periodic.

6.3 Hajós Groups

In 1941, G. Hajós solved a famous problem of Minkowski in rephras-

ing it into a problem of factoring a finite Abelian group into certain
subsets. Minkowski wanted to show that in any simple lattice, tiling
of the n-dimensional Euclidean space in unit cubes must share a com-
plete (n − 1)-dimensional face. Latter, around 1948, Hajós thought
that in a factorization of ZN in two factors, one of the factors had
to be periodic. In fact, this conjecture is false, and counterexamples
are precisely Vuza canons.

Definition 6.7. Let G be a finite cyclic group. A nonempty set A

is periodic if there exists an element g of G such that A + g = A
 Rhythmic Canons 163

and g = 0 (the identity element of G). Otherwise, the set A is called

aperiodic or nonperiodic set.
Definition 6.8. Let G be a finite cyclic group. G is called a Hajós
group if for each factorization of G = A ⊕ B at least one factor is
periodic.
Remark 6.3. If G is a finite Abelian group, the factors are not
necessary subgroups of G. For example, if A = {0, 1, 4, 5} and
B = {0, 2}, A ⊕ B = Z8 The set A is 4-periodic, but none of A
and B are subgroups of Z8 .
A factorization of a cyclic group G is said to be trivial if at least
one of its factors is periodic. If every factorization of G is trivial, G is
called a “good group”. Otherwise, if no factor is periodic, G is called
a “bad group”. In 1950, Hajós (1950a, 1950b) proved that there exist
cyclic groups admitting a nontrivial factorization.
Proposition 6.11. Let G be a finite cyclic group and A, B subsets
of G. If A is g-periodic, then A + B is also periodic with period g.
Proof. This results from the following computation:
g + (A + B) = (g + A) + B = A + B 
Remark 6.4. If A+ B is periodic, then it does not follow that either
A or B is periodic.
de Bruijn (1953a) showed that if N = d1 d2 d3 where d3 > 1, d1 and
d2 are coprime and if both d1 and d2 are composite numbers, then the
cyclic group ZN of order N is bad. In de Bruijn (1953b), de Bruijn
remarked that the cyclic groups which are not covered by this result
are those of orders pα (α ≥ 1), pα q (α ≥ 1), p2 q 2 , pα qr (α = 1, 2),
pqrs where p, q, r, s denote different primes. Redéi (1950) proved that
the cyclic groups of orders pα (α ≥ 1), pq, pqr are “good”. de Bruijn
(1953b) showed that pα q (α ≥ 1) are “good groups”. The remaining
cases p2 q 2 , p2 qr and pqrs were showed by Sands (1957) in 1957. In
1962, Sands gave the classification of all finite Abelian groups which
are Hajós groups (Sands, 1962, 2004).
Theorem 6.4 (Sands). The group ZN is an Hajós group if and
only if N is of the form pk for k ≥ 0, pk q for k ≥ 1, p2 q 2 , pqr, p2 qr
or pqrs for distinct primes p, q, r, s.
164 A Compendium of Musical Mathematics

Theorem 6.5 (de Bruijn). The group ZN is a non-Hajós group if

and only if N can be expressed in the form p1 p2 n1 n2 n3 , where p1 , p2
are primes, pi ni ≥ 2 for i = 1, 2, 3 and gcd(n1 p1 , n2 p2 ) = 1.
A computation shows that the smallest values of N for which ZN
is a non-Hajós group are as follows: 72, 108, 120, 144, 168, 180, 200,
216, 240, 252, 264, 270, 280, 288, 300, 312, 324, 336, 360, 378, 392,
396, 400, 408, 432, 440, 450, 456, 468, 480, 500, 504, 520, 528, 540,
552, 560, 576, 588, 594, 600, 612, 616, 624, 648, 672, 675, 680, 684,
696, 700, 702, 720, 728, 744, 750, 756, 760, 784, 792, 800, 810, 816,
828, 864, 880, 882, 888, 900, 912, 918, 920, 936, 945, 952, 960, 968,
972, 980, 984, 1000, etc.

6.4 Coven–Meyerowitz Conjecture

The following results have been established by Ethan Coven and

Aaron Meyerowitz (1999). We summarize the main results.
Theorem 6.6. Let A be a finite set of nonnegative integers and let
SA be the set of prime powers s such that the cyclotomic polynomial
Φs (x) divides themask polynomial A(x) of A. If
(T1 ) A(1) = Φs (1) = A(1) = |A|
s∈SA
(T2 ) if s1 , s2 , . . . , sm ∈ SA are powers of distinct primes, then
Φs1 ···sm (x) divides A(x),
then A tiles.
Theorem 6.7. Under the same hypothesis, if A tiles, then (T1 ) is
true.
Theorem 6.8. Under the same hypothesis, if |A| has at most two
different prime factors and A tiles, then both (T1 ) and (T2 ) are true.
In general, it is not known whether the conditions (T 2) are always
necessary for tiling.
Conjecture 6.1 (Coven–Meyerowitz). Under the same hypoth-
esis, A tiles the integers if and only if (T1 ) and (T2 ) are true.
 Rhythmic Canons 165

Proposition 6.12. Let (A, B) be a rhythmic canon, A ⊕ B = ZN .

Consider the set
RA = {d, Φd (x)|A(x)} and SA = {pα ∈ RA , p is prime, α ∈ Z> }
Then SA , SB are disjoint sets whose union is the set of all prime
powers dividing N, and RA ∪ RB is the set of all divisors of N
excepted {1}.
Example 6.6. For N = 72, the sets A = {0, 18}⊕{0, 8, 16} and B =
{0, 1, 5, 6, 12, 25, 29, 36, 42, 48, 49, 53} form a rhythmic canon. Their
mask polynomials are A(x) = Φ3 Φ4 Φ6 Φ212 Φ24 Φ36 (x) and B(x) =
Φ2 Φ8 Φ9 Φ18 Φ72 ψ(x). Hence,
RA = {3, 4, 6, 12, 24, 36} and RB = {2, 8, 9, 18, 72}
and
SA = {3, 4} and SB = {2, 8, 9}

6.5 Fuglede Conjecture

In 1974, Bengt Fuglede formulated the conjecture that a Borel set of

finite positive Lebesgue measure in Rd is spectral if and only if it is a
tile. In 2004, Terence Tao showed that the conjecture fails for d ≥ 5,
and two years later, Kolountzakis, Matolcsi and Farkas showed that
it is false for d ≥ 3. But the cases d = 1 and d = 2 which are linked
with rhythmic canons are still open.
Let G be a locally compact Abelian group and G  its dual group.
Let Ω be a subset of G of finite nonzero Haar measure.
Definition 6.9. The set Λ is a spectrum of Ω if the characters (λ)λ∈Λ
form an orthonormal basis in L2 (Ω). Then Ω is called a spectral set.
The Fuglede conjecture can be stated as follows:
Conjecture 6.2. The set Ω ∈ G is spectral if and only if Ω is a tile.
It has been stated that the conjecture is true for some groups, such
as Zp × Zp with p prime by Iosevich, Mayeli and Pakianathan in 2015
166 A Compendium of Musical Mathematics

and Zpnq with p, q distinct primes by Malikiosis and Kolountzakis in

2016. And for the cyclic group ZN , for some values of N : N = pn q by
Kolountzakis in 2017, N = Zpqr with p, q, r distinct prime numbers
by Ruxi Shi in 2019, for N = p2 qr by Somlai in 2023.
Using the abbreviations
T-S(G): in G, Tile =⇒ Spectral
S-T(G): in G, Spectral =⇒ Tile
we have the following result, which underlines the importance of the
cyclic group in this question.
Theorem 6.9 (Dulkay–Lai, 2014). The following sentences hold:
(1) T-S(R) ⇐⇒ T-S(Z) ⇐⇒ T-S(ZN ), ∀N ∈ N.
(2) S-T(R) =⇒ S-T(Z) =⇒ S-T(ZN ), ∀N ∈ N.
Let A be a subset of ZN . The mask polynomial is the Fourier
transform of the indicator function of A. Denoting ζN = e2iπ/N the
N -th root of unity, we have for all n ∈ ZN

−an −n
F (1A ) (n) = ζN = A(ζN )
a∈A

Let ZA be the set of the zeros of the mask polynomial

ZA = {n ∈ ZN , A(ζN
n
) = 0}
Proposition 6.13. n ∈ ZA if and only if nr ∈ ZA , for all r ∈
ZN \{0}.
Definition 6.10. An n × n complex matrix H is a complex
Hadamard matrix if all entries of H have absolute value 1 and
HH ∗ = nI with I is the identity matrix.
That is to say that the rows (and the columns) of H form an
orthogonal basis of Cn . In other words, all entries of H are n-th
roots of unity whose rows are mutually orthogonal.
Definition 6.11. A subset A ⊆ ZN is spectral if there is a set B ⊂
ba , b ∈ B} forms an orthogonal basis of L2 (A).
ZN such that the set {ζN
B is called the spectrum of A and (A, B) a spectral pair.
 Rhythmic Canons 167

Proposition 6.14. (A, B) is a spectral pair if and only if |A| = |B|

and
 (b−b )a
ζN = 0, ∀b, b ∈ B, b = b
a∈A

Proposition 6.15. (A, B) is a spectral pair if and only if the matrix

 ba  T
H = ζN a∈A,b∈B
is a complex Hadamard matrix M M = |A| I.

Theorem 6.10. Let A, B ⊆ ZN . The following statements are

equivalent:

(1) The pair (A, B) is spectral.

(2) |A| = |B| and (B\B)\{0} ⊂ ZA .
(3) The matrix H = (e2iπab/N )a,b is a complex Hadamard matrix.

6.6 Vuza Canons

Vuza canons were introduced by Vuza under the name of “Regular

Complementary Canons of Maximal Category” in his seminal paper
(Vuza, 1991). We adopt the following general definition.

Definition 6.12. Let G be a finite cyclic group and A, B two

nonempty subsets of G. A (generalized) Vuza canon (A, B) of G
is a rhythmic canon

G=A⊕B

where neither A nor B is periodic.

Vuza gave the first musical application of the factorization of

cyclic groups and computed the first “Vuza canon” that appears for
N = 72. Today, several composers have taken an interest in Vuza’s
canons, such as Sébastien Roux in Musiques d’ordinateur or Fabien
Lévy in Soliloque (1999), Coı̈ncidences (2001), see Lévy (2011).
Moreover, Vuza showed the following equivalence of the conditions
on order N (Vuza, 1991), which have been stated by Sands and de
Bruijn.
168 A Compendium of Musical Mathematics

Theorem 6.11. For every integer N ≥ 2, the following statements

are equivalent:
(1) Vuza canons exist in ZN .
(2) N = n1 n2 n3 p1 p2 where p1 , p2 denote different primes, pi ni ≥ 2,
for i = 1, 2 and n1 p1 and n2 p2 are relatively prime.
(3) N is not one of the following:
pα (α ≥ 1), pα q (α ≥ 1), p2 q 2 , pα qr (α = 1, 2), pqrs
where p, q, r, s are different primes.
The following lemma is shown in Szabo and Sands (2009).
Lemma 6.2. Let G be a finite cyclic group, H a subgroup of G and
A, B two subsets of G such that A ⊆ H, then
(A + B) ∩ H = A + (B ∩ H)
Proof.
(1) Let g be an element of A + (B ∩ H), that is, g = a + b with
a ∈ A and b ∈ B ∩ H. Since g is also an element of A + B and
an element of H, we have the inclusion
A + (B ∩ H) ⊆ (A + B) ∩ H
(2) Conversely, given g ∈ (A+B)∩H, g belongs to A+B and g ∈ H.
Since g ∈ A+ B, there exist a ∈ A and b ∈ B such that g = a+ b.
But g is in H and a ∈ A ⊆ H. Thus, b ∈ H as H is a subgroup
of G. Thus, the element b is in B ∩ H and so
(A + B) ∩ H ⊆ A + (B ∩ H)

Theorem 6.12. Let G be a finite cyclic group and H a subgroup of
G. If (A, B) is a Vuza canon of G and if A is a subset of H, (B ∩ H)
is nonperiodic in H, then (A, B ∩ H) is a Vuza canon of H.
Proof. The equations
H = G ∩ H = (A + B) ∩ H = A + (B ∩ H)
show that H is the sum of A and B ∩ H, as A is a subset of H. This
sum is direct since the sum A + B is. Moreover, suppose that A is
periodic in H. Since A ⊂ H ⊂ G, the set A is periodic in G. And
this leads to a contradiction. 
 Rhythmic Canons 169

Corollary 6.3. Let k be a positive integer and N a Vuza order. If

(A, B) is a Vuza canon of ZkN and A and (B ∩ kIN ) two nonperiodic
subsets of kIN , then the pair (A , B  ) defined by

A = A/k, B  = (B ∩ kIN )/k

is a Vuza canon of ZN .
Proof. Since kIN can be seen as a subgroup of ZkN , we can apply
the previous theorem. Thus, (A, B ∩ kIN ) is a Vuza canon of kZN .
Taking the quotient by k, since A , B  are in ZN , (A , B  ) is a Vuza
canon of ZN . 

Example 6.7. For instance, consider the Vuza canon A = {0, 16,
32, 36, 52, 68} and

B = {0, 7, 12, 15, 24, 33, 34, 45, 46, 55, 57, 58, 63, 72, 84, 96, 103,
105, 106, 111, 117, 118, 129, 130}

of ZN k = Z144 with N = 72 and k = 2. Since H = 2Z72 is a subgroup

of Z144 , the pair

A = A/2 = {0, 8, 16, 18, 26, 34}

B  = (B ∩ 2I72 )/2 = {0, 6, 12, 17, 23, 29, 36, 42, 48, 53, 59, 65}

is a Vuza canon of Z72 .

Theorem 6.13. Let N = n1 n2 n3 p1 p2 ∈ N× where p1 , p2 denote
different primes, pi ni ≥ 2, for i = 1, 2 and gcd(n1 p1 , n2 p2 ) = 1.
Denote K1 and K2 the sets

K1 = n2 n3 (Ip2 ⊕ p2 n1 Ip1 )
K2 = n1 n3 (Ip1 ⊕ p1 n2 Ip2 )

and let Tj (K2 ) = {j} ⊕ K2 be the translation of K2 by j. The canon

(A, B) of lengths |A| = n1 n2 and |B| = p1 p2 n3 defined by

A = n3 (p2 n2 In1 ⊕ p1 n1 In2 )

B = K1 ∪ T1 (K2 ) ∪ · · · ∪ Tn3 −1 (K2 )

is a Vuza canon.
170 A Compendium of Musical Mathematics

Proof. The proof uses the following properties:

Ia ⊕ aIb = Iab
cIa ⊕ aIbc = Iabc mod abc gcd(a, c) = 1

Compute the direct sum of A with K1

A ⊕ K1 = p2 n2 n3 In1 ⊕ p1 n1 n3 In2 ⊕ n2 n3 Ip2 ⊕ p2 n1 n2 n3 Ip1

= n2 n3 (p2 In1 ⊕ Ip2 ) ⊕ n1 n3 (p1 In2 ⊕ p2 n2 Ip1 )
= n2 n3 In1 p2 ⊕ n1 n3 (p1 In2 ⊕ p2 n2 Ip1 )
= n2 n3 (In1 p2 ⊕ p2 n1 Ip1 ) ⊕ n1 n3 p1 In2
= n2 n3 Ip1 p2 n1 ⊕ n1 n3 p1 In2
= n3 (n1 p1 In2 ⊕ n2 Ip1 p2 n1 )
= n3 Ip1 p2 n1 n2

The direct sum of A with K2 ,

A ⊕ K2 = p2 n2 n3 In1 ⊕ p1 n1 n3 In2 ⊕ n1 n3 Ip1 ⊕ p1 n1 n2 n3 Ip2

= p1 n1 n3 (In2 ⊕ n2 Ip2 ) ⊕ n3 (p2 n2 In1 ⊕ n1 Ip1 )
= n1 n3 (Ip1 ⊕ p1 In2 p2 ) ⊕ n2 n3 p2 In1
= n3 (n1 Ip1 p2 n2 ⊕ p2 n2 In1 )
= n3 Ip1 p2 n1 n2

Lastly, the set A ⊕ B tiles the set IN

A ⊕ B = n3 Ip1 p2 n1 n2 ∪ {1} ⊕ n3 Ip1 p2 n1 n2 ∪ · · · ∪ {n3 − 1}

⊕n3 Ip1 p2 n1 n2
= Ip1 p2 n1 n2 n3
= IN

 Rhythmic Canons 171

Example 6.8. For n = 72, Sands–de Bruijn decomposition is p1 =

n1 = n3 = 2 and p2 = n2 = 3. The previous result leads to K1 = 6I3 ⊕
36I2 = {0, 6, 12, 36, 42, 48}, K2 = 4I2 ⊕ 24I3 = {0, 4, 24, 28, 48, 52}.
The computation of

B1 = K1 ∪ T1 (K2 ) = {0, 1, 5, 6, 12, 25, 29, 36, 42, 48, 49, 53}

and

A1 = 18I2 ⊕ 8I3 = {0, 8, 16, 18, 26, 34}

gives the Vuza canon (A1 , B1 ) for the non-Hajós group Z72 . There
are three nonisomorphic solutions for the inner rhythm, namely A1 =
{0, 18}⊕{0, 8, 16}, A2 = {0, 18}⊕{0, 8, 40}, A3 = {0, 18}⊕{0, 8, 64},
but there is only one nonisomorphic solution since A2 = 5A1 and
A3 = 7A1 .

The following results have been set in Jedrzejewski (2013) and

improved by Lanzarotto and Pernazza (Lanzarotto, 2021; Lanzarotto
and Pernazza, 2022). The proof of the aperiodicity of the set A ⊂ ZN
follows from the following result:

Proposition 6.16. A set A ⊂ ZN is periodic of period k | N if and

only if

xN − 1
| A(x)
xk − 1

Proof. We have

A + k = A ⇔ xk A(x) = A(x) mod (xN − 1)

⇔ xN − 1 | (xk − 1)A(x)
xN − 1
⇔ | A(x)
xk − 1

172 A Compendium of Musical Mathematics

Remark 6.5. From this proposition follows that a set A ⊂ ZN is

aperiodic iff for all k | N, N = k,

xN − 1
 A(x)
xk − 1
in other words iff for all divisors of N , different of N , there exist a
divisor d of N which is not a divisor of k such that the cyclotomic
polynomial Φd (x)  A(x).
Theorem 6.14. Let N = n1 n2 n3 p1 p2 ∈ N where n1 , n2 , n3 , p1 ,
p2 > 1, and gcd(n1 p1 , n2 p2 ) = 1. Let us denote

A1 = n1 p1 n3 In2 A2 = n2 p2 n3 In1
U1 = n1 n2 n3 p1 Ip2 U2 = n1 n2 n3 p2 Ip1
V1 = n2 n3 Ip2 V2 = n1 n3 Ip1
K1 = {0} K2 = {1, 2, . . . , n3 − 1}

then the pair (A, B) defined by

A = A1 ⊕ A2
B = (U1 ⊕ V2 ⊕ K1 )  (U2 ⊕ V1 ⊕ K2 )

where  denotes the disjoint union is a Vuza canon of ZN .

Proof. (1) We start to show that

A1 ⊕ U1 ⊕ V2 = A1 ⊕ U1 ⊕ V1

We have

A1 ⊕ U1 ⊕ V2 = n1 p1 n3 In2 p2 ⊕ n1 n3 Ip1
= n1 n3 (Ip1 ⊕ p1 In2 p2 )
= n1 n3 In2 p1 p2

In the same way,

A1 ⊕ U1 ⊕ V1 = n1 p1 n3 In2 p2 ⊕ n1 n2 n3 p2 Ip1
= n1 n3 (p1 In2 p2 ⊕ n2 p2 Ip1 )
= n1 n3 In2 p1 p2
 Rhythmic Canons 173

Thus, A1 ⊕ U1 ⊕ V2 = A1 ⊕ U1 ⊕ V1 . The same expression can be set

for A2 :
A2 ⊕ U2 ⊕ V1 = n2 p2 n3 In1 p1 ⊕ n2 n3 Ip2
= n2 n3 (p2 In1 p1 ⊕ Ip2 )
= n2 n3 In1 p1 p2
and
A2 ⊕ U2 ⊕ V1 = n2 p2 n3 In1 p1 ⊕ n1 n2 n3 p1 Ip2
= n2 n3 (p2 In1 p1 ⊕ n1 p1 Ip2 )
= n2 n3 In1 p1 p2
and consequently,
A2 ⊕ U2 ⊕ V1 = A2 ⊕ U2 ⊕ V1
(2) Using these relations, we can know compute the sum

A ⊕ B = (A1 ⊕ A2 ) ⊕ ((U1 ⊕ V1 ⊕ K1 )  (V1 ⊕ U2 ⊕ K2 ))

= (A1 ⊕ A2 ⊕ U1 ⊕ V2 ⊕ K1 )  (A1 ⊕ A2 ⊕ V1 ⊕ U2 ⊕ K2 )
= (A1 ⊕ A2 ⊕ U1 ⊕ U2 ⊕ K1 )  (A1 ⊕ A2 ⊕ U1 ⊕ U2 ⊕ K2 )
= A1 ⊕ A2 ⊕ U1 ⊕ U2 ⊕ (K1  K2 )
= (A1 ⊕ U1 ) ⊕ (A2 ⊕ U2 ) ⊕ K
But
A1 ⊕ U1 = n1 n3 p1 In2 p2
A2 ⊕ U2 = n2 n3 p2 In1 p1
it follows that
(A1 ⊕ U1 ) ⊕ (A2 ⊕ U2 ) = n3 In1 p1 n2 p2
and the sum is equal to the cyclic group:
A ⊕ B = (A1 ⊕ U1 ) ⊕ (A2 ⊕ U2 ) + K
= n3 In1 p1 n2 p2 + In3 = ZN
(3) It remains to prove that A and B are aperiodic. As suggest in
Lanzarotto (2021), we use Remark 6.5. Fixing any k a divisor of N ,
174 A Compendium of Musical Mathematics

different of N , we look for a divisor d of N which is not a divisor

of k such that the cyclotomic polynomial Φd (x)  A(x), the mask
polynomial:
 n1 n2 n3 p 1   n1 n2 n3 p 2 
x −1 x −1
A(x) =
xn 1 n 3 p 1 − 1 xn 2 n 3 p 2 − 1
They are two cases (1) n1 n3 p1  k and Φn1 n3 p1 (x)  A(x) and (2)
n2 n3 p2  k and Φn2 n3 p2 (x)  A(x). Because if we had n1 n3 p1 | k
and n2 n3 p2 | k, then k would be k = an1 n3 p1 = bn2 n3 p2 and
an1 p1 = bn2 p2 . But since gcd(n1 p1 , n2 p2 ) = 1, it follows that
a = n2 p2 and b = n1 p1 and thus k = N , which contradicts the
hypothesis. Consequently, A is aperiodic. The same is done for B.

The enumeration of Vuza canons is still an open problem.

6.7 Exercises

Exercise 6.1
Let N = 16 and A = {0, 5, 8, 13}. Write down the mask polynomial of
A(x) and its decomposition into cyclotomic polynomials. Find a set
B such that A⊕B = ZN . Using Coven–Meyerowitz (CM) conditions,
prove that A tiles Z16 .
Exercise 6.2
Let us consider the rhythmic canon A ⊕ B = ZN , k ∈ N× and the
set

C = A ⊕ {0, N, 2N, . . . , (k − 1)N }

(1) Prove that C ⊕ B = ZkN or, in other words, that the concatena-
tion k times of A tiles ZkN with the same tile B. (2) Prove that the
CM condition (T2 ) is stable by concatenation, that is, if A tiles ZN
and satisfies (T2 ), then the set C verifies (T2 ).
Exercise 6.3
Prove that the CM condition (T1 ) is not sufficient for A to tiles. Hint:
Consider the set A = {0, 1, 3, 4, 6, 7}.
 Rhythmic Canons 175

Exercise 6.4
Let A ⊂ N be finite and n ∈ N. Let us consider B = A + n. (1) Prove
that A satisfies (T1 ) iff B satisfies (T1 ). (2) Prove that A satisfies
(T2 ) iff B satisfies (T2 ).

Exercise 6.5
Let A ⊂ N be finite and n ∈ N. Let us consider B = nA. (1) Prove
that A satisfies (T1 ) iff B satisfies (T1 ). (2) Prove that if n = p is
prime, A satisfies (T2 ) iff B satisfies (T2 ).

References

Amiot, E. (2005). Rhythmic canons and Galois theory, Colloquium on Mathemat-

ical Music Theory, Grazer Mathematische Berichte 347, 1–21.
Amiot, E. (2011). Structures, algorithms, and algebraic tools for rhythmic canons,
Perspectives of New Music 49(2), 93–142.
Coven, E. and Meyerowitz, A. (1999). Tiling the integers with translates of one
finite set, Journal of Algebra 212(1), 161–174.
Davalan, J.P. (2011). Perfect rhythmic tilings, Perspectives of New Music 49(2),
144–197.
de Bruijn, N.G. (1953a). On the factorization of finite Abelian groups, Indaga-
tiones Mathematicae 15, 258–264.
de Bruijn, N.G. (1953b). On the factorization of cyclic groups, Indagationes Math-
ematicae 15, 370–377.
Fripertinger, H. (2001). Enumeration of nonisomorphic canons, Tatra Mountains
Mathematical Publications 23, 47–57.
Hajós, H. (1950a). Sur la factorization des groupes abéliens, Casopis Pro Pesto-
vani Matematiky A Fysiky 74, 157–162.
Hajós, H. (1950b). Sur le probleme de factorization des groupes cycliques, Acta
Mathematica Academiae Scientiarum Hungarica 1, 189–195.
Jedrzejewski, F. (2009). Tiling the integers with aperiodic tiles, Journal of Math-
ematics and Music 3(2), 99–115.
Jedrzejewski, F. (2013). Enumeration of Vuza canons. arXiv preprint 304.6609.
Lanzarotto, G. (2021). Extended Vuza Canons, Ph.D. dissertation. Pernazza L.
Université Degli Studi Di Pavia E Di Milano-Bicocca.
Lanzarotto, G. and Pernazza, L. (2022). Extended Vuza canons. In: Interna-
tional Conference on Mathematics and Computation in Music, Springer,
112–126.
Lévy, F. (2011). Three uses of Vuza canons, Perspectives of New Music 49(2),
23–31.
176 A Compendium of Musical Mathematics

Redéi, L. (1950). Ein Beitrag zum Problem der Faktorisation von endlichen
Abselschen Gruppen, Acta Mathematica Academiae Scientiarum
Hungarica 1, 197–207.
Sands, A.D. (1957). On the factorisation of finite Abelian groups. I, Acta
Mathematica Hungarica 8, 65–86.
Sands, A.D. (1962). On the factorisation of finite Abelian groups. II, Acta
Mathematica Hungarica 13, 153–169.
Sands, A.D. (2004). Factoring finite Abelian groups, Journal of Algebra 275,
540–549.
Szabo, S. and Sands, A.D. (2009). Factoring Groups into Subsets, CRC Press,
Boca Raton.
Vuza, D.T. (1991). Supplementary sets and regular complementary unending
canons, Perspectives of New Music 29(2), 22–49; 30(1), 184–207; 30(2),
102–125; 31(1), 270–305.
 Chapter 7

Serial Knots

The 12-tone technique, also known as dodecaphony, is a method of

musical composition devised by Arnold Schönberg in 1921 and used
for the first time in his fifth Piano pieces op. 23. At the same time,
or maybe previously, Josef Matthias Hauer has developed a simi-
lar technique based on tropes of six notes. The 12-tone technique is
based on an ordered arrangement of the 12 notes of the chromatic
scale called the tone row or the series (from there comes the name
of serial music). This technique spread and generalized during the
second quarter of the 20th century. The story is told in many books,
including Jedrzejewski (2021). In this chapter, we show how the
12-tone rows are classified according to chord diagrams of singular
knots with six double points and we give an overview of some theories
associated with serial music.

7.1 Chord Diagrams

Chord diagrams (sometimes called Gauss diagrams) are well known

in low-dimensional topology as an important tool for studying knots.
We use them to classify 12-tone rows.
Each 12-tone row, for instance,

C, A, G, G, D, E, A, B, D, F, E , B 

is written, according to the identification of Z12 , in the compact form

0867249e153t with e (eleven, 11) and t (ten, 10).

177
178 A Compendium of Musical Mathematics

Before starting to compose with a prime tone row (P), musicians

build a table of at most 48 interval preserving forms derived from the
series, named inversion (denoted I), retrograde (R) or retrograde-
inversion (RI). The tone row can be transposed 12 times, reading
from left to right (P), or in the reverse order, from right to left,
leading to 12 new forms (R). It can also be inversed and trans-
posed, reading from left to right (I) or, in the reverse order, from
right to left (RI). So we get up to 48 different row forms. Some
rows have fewer due to symmetry. For instance, the row of Alban
Berg’s Schliesse mir dir Augen beide 0e742938t156 transposed at the
tritone (651t819247e0) is the same as the retrograde (P6 = RP0 ), and
relation is valid for all transpositions (Pn+6 = RPn for all n). This
symmetry is the reason why there are less than 48 derived forms.
A simple way to represent a tone row is to build its chord diagram.
For instance, consider the tone row 1754t90e2836, put the notes of
the series on a circle and join the tritones together by a chord (see
Fig. 7.1). Then rub out notes. You get an abstract structure called a
chord diagram.
It is easy to verify that the 48 derived forms of the 12-tone row
are mapped on the same chord diagram. All the transpositions of the
series have the same chord diagram (up to a rotation of the diagram).
The retrogradation of the tone row is placed on the diagram which
is the mirror symmetry of the original diagram. The inversion of the
series uses the same chord diagram and the retrogradation of the
inversion is mapped by the mirror diagram. All the 48 derived serial

Fig. 7.1. Chord diagram D155 of the tone row 1754t90e2836.

 Serial Knots 179

forms use the same chord diagram, viewed in the space, possibly
mirroring or rotating. That is why chord diagram can be used to
classify 12-tone rows.

7.2 Enumeration of Tone Rows

The enumeration of tone rows and their mathematical properties has

been studied extensively by Harald Fripertinger (2015) and Robert
Morris (1977, 1985, 1988, 1989, 2007). David Reiner (1985) was the
first to enumerate the 12-tone rows according to the equivalence of
the 48 derived forms.
Theorem 7.1 (Reiner, 1985). The number of classes of 12-tone
rows equivalent under transposition, retrograde and inversion is
9, 985, 920
Proof. Let X be the set of tone rows and S the set of classes of
12-tone rows equivalent under transposition, retrograde and
inversion. The cardinal of X is |X| = 12!. Let R be the retrograde
transformation, and let T and I be the transposition and the inver-
sion. The problem of counting the tone row of S is equivalent to
count the number of orbits of the group G = D12 × Z2 generated
by the transformations T , R and I, on the set X. D12 denotes the
dihedral group of order 24. Burnside’s lemma says that the number
of orbits is the average number of fixed points, namely
1 
|S| = |Xg |
|G|
g∈G

Since G has 48 elements, we have to determine the fixed points for

each element. But the elements T n (1 ≤ n ≤ 11), T n R (n = 6), T n I,
T n IR (for n even) have no fixed points and the identity has 12! fixed
points. It remains to look at seven elements of G, the operation T 6 R
and T n IR for n odd. The fixed tone rows by the operation T 6 R are
the tone rows in which the six last entries are the reverse of the six
first transposed by a tritone. So there are
t = 12 × 10 × 8 × 6 × 4 × 2 = 26 · 6!
ways of doing this. The same can be done for the transformation
T n IR: the first six entries determine the number of fixed tone rows.
180 A Compendium of Musical Mathematics

So they are also 26 · 6! series, for each n odd. Finally, there are

12! + 7 · 26 · 6!

fixed tone rows. Thus, we get |S| = (12! + 7t)/48 = 9985920 tone row
classes. 
More generally, for the N -EDO, we have the following result,
which is a direct application of the de Bruijn extension of the Pólya
enumeration theorem.
Theorem 7.2. In the N -EDO (N ≥ 3), there are exactly N !
tone rows of N notes. Under the equivalence of the derived forms
(i.e. under the action of the dihedral group), there are
  
1 (N −4)/2 1 N + 1 ! if N is even
4 (N − 1)! + 2 N 2
1
 
(N −5)/2 N −1 !
4 (N − 1)! + 2 2 if N is odd

tone rows of N notes.

The enumeration of chord diagrams has been set up by Khruzin
(2000).
Theorem 7.3 (Khruzin, 2000). Under the action of the cyclic
group C2n , the number of equivalent diagrams is
1 
cn = ϕ(i)νn (i)
2n
i|2n

where ϕ(i) is Euler’s function and νn is defined for all the divisors
of 2n by
⎧ n/i
⎨ i (2n/i − 1)!!
⎪ if i is odd
νn (i) =  ni  2n/i
⎪
⎩ ik (2k − 1)!! if i is even
k=0 2k

Under the action of the dihedral group, the number of Gauss diagrams
in the 2n-EDO is given by
1 1
dn = cn + (κn−1 + κn )
2 2
 Serial Knots 181

with

2
n

n!
κn =
k!(n − 2k)!
k=0

where x indicates the greatest integer less than or equal to x.

The following table gives the number of chord diagrams (dn ) for
the 2n-EDO. For the common equal temperament (n = 6), there
are exactly 554 chord diagrams which classify all 12 tone rows with
respect to their structural properties.
2n-EDO cn dn
6 5 5
8 18 17
10 105 79
12 902 554
14 9749 5283
16 127072 65346
18 1915951 966156
20 32743182 16411700
22 624999093 312700297

A complete catalog of the 554 chord diagrams can been found

in Jedrzejewski (2021). Diagrams are classified by the Gauss word
order. Each of the six chords are labeled by the numbers 1, 2, . . . , 6
and the Gauss word is a permutation of the word 112233445566. To
construct the Gaussian word for a chord diagram, one has to choose a
starting point on the circle, a direction of rotation of the circle, then
assign to each of the 12 points of the circle, in the order in which the
chords are met, a label with the numbers 1 to 6. A single chord has
the same number at both ends. The Gaussian word is the minimum
number that you can form by this procedure by varying the starting
point and the direction of travel on the circle. In other word, two
Gauss words over the same alphabet are isomorphic if one can be
obtained from the other by cyclic shifts and reversing the word.
Definition 7.1. For two chord diagrams D and L, we said that D
is less than L (D < L) if and only if w(D) < w(L), where w(D) is
the Gauss word of D.
182 A Compendium of Musical Mathematics

With the Gauss order, the 554 serial chord diagrams can be
classified. The Gauss word can be written in a form slightly dif-
ferent by locating the chords of different types: the chord a connects
two consecutive points of the Gauss diagram, b connects the two
extreme points of three consecutive points of the circle, and so on,
until the type f which connects the two extreme points of seven con-
secutive points of the circle. Moreover, one uses the opposite (e.g.
e−1 is a chord linking the two extreme points of eight consecutive
points). For example, the series of Jean Barraqué Au-delà du hasard
087142t3e569 is represented by the word
X = abd−1 dab
The structural vector of a chord diagram is made of six compo-
nents which count the number of chords of each type a, b, c, d, e, f .
For instance, the structural vector of the previous chord diagram is
220200. This means that the diagram includes two chords of type a,
two chords of type b, none of type c, two chords of type d and none
of types e and f .
The intertwining properties of chords in chord diagrams, found in
the Gauss or structural word, are very important for the analysis of
musical serial structures, as well as indexing under the same chord
diagram.
Example 7.1. The chord diagram used by Karel Goeyvaerts
in his Sonata for two pianos (123679t540e8) has a Gauss word
121324546356, a structural word bcf bdc and a structural vector
022101.
A statistical study of 1132 12-tone rows (see Jedrzejewski, 2021)
has shown that the diagram most used by composers is the diagram
of the Symphony, op. 21 from Webern (5876t934012e).

7.3 All-Interval 12-Tone Rows

All-interval 12-tone rows have been studied by André Riotte (1963,

2006), Herbert Eimert (1963) and Eduard Herzog (1964) who have
published at the same time a complete list of the 1928 all-interval
tone rows. These tone rows can be classified under 63 chord dia-
grams. The first known all-interval row (5409728136te) was named
 Serial Knots 183

the Mutterakkord by Fritz Heinrich Klein, who created it in 1921 in

his chamber-orchestra composition Die Maschine.
The chord diagram of all-interval tone rows includes two (and
only two) chords of type a because the first and the last notes of the
series are perforce a tritone. Remark that all types of chords are not
necessarily present in an all-interval tone row. For example, in the
Symphonia brevis op. 16 of Hanns Jelinek (392481te7506), the chord
diagram has a structural word a2 bc2 b, in which there are no chords
of type d, e or f .
The 1928 all-interval tone rows are indexed to 63 chord
diagrams. They therefore represent approximately 11% of serial
structures (63/554 chord diagrams). They are thus more visible than
in the classification by their interval structure, since the 1928 all-
interval tone rows are embedded in the middle of the 9 985 920 tone
rows (counted up to transposition and inversion) and then only rep-
resent 0.02%.
Three chord diagrams cover more all-interval tone rows than the
others: the diagram of Elliott Carter’s Caténaires (62e9134t5870)
indexes 96 tone rows, the diagram of Webern’s Symphony op. 21
indexes 88 tone rows and the diagram of the third String Quartet
(0e295348t716) of Elliott Carter indexes 80 tone rows.
In the following are two chord diagrams D74 (e2658143t907,
Bernstein’s Kaddish Symphony) and D154 (39452e18t607, Skalkot-
tas’s Violin Concerto AK22) each index 64 tone rows. The D50
chord diagram used, for instance, in Milton Babbitt’s Solo Requiem
(02375t4198e6) indexes 56 tone rows. Then 9 chord diagrams index
48 tone rows, 3 diagrams 40 tone rows, 16 diagrams 32 tone rows,
1 diagram 24 tone rows, 21 diagrams 16 tone rows and 7 diagrams 8
tone rows. Beyond the number 358, no more chord diagram indexes
all-interval 12-tone rows. See Jedrzejewski (2021) for details.

7.4 Types of Tone Rows

The classification of the 12-tone rows proposed by Edmond Costère

(1962) and formalized by Bernard Parzysz (1984) is a classification
based on the division of the tone row S in two parts. The set of the
six first notes of the series is denoted A and the set of the six last
notes is denoted Ac , the complement of A in the 12 notes, Z12 .
184 A Compendium of Musical Mathematics

Definition 7.2. A transpose tone row is a tone row in which there

is a transposition from the set A to the set Ac . Denote the set of the
transpose tone row by T , we have
S ∈ T ⇔ ∃ i Ti (A) = Ac
Example 7.2. The 12-tone row of Arnold Schönberg’s Serenade
opus 24 (423e0186957t) which has the chord diagram D301 and the
tone row of B.A. Zimmermann’s Die Soldaten (9561428t70e3) which
has the chord diagram D358 are transpose tone rows because in each
case, the transposition of the set A gives the set Ac (T6 (A) = Ac ).
Definition 7.3. A reverse tone row is a tone row for which it exists
a transposition from the set I(A) = −A to the set Ac . Denote R the
set of reverse series, we have
S∈R⇔∃i Ti (−A) = Ac
Example 7.3. The 12-tone row of Luigi Dallapicolla’s Il Prigioniero
(8e275t934016) which has the chord diagram D238 is a reverse tone
row because we have T11 (−A) = Ac . The tone row of Pierre Boulez’s
Structures for two pianos (329876410t5e) is also a reverse tone row,
as we have T7 (−A) = Ac .
Definition 7.4. A semi-transpose tone row is a tone row for which
it exists a transposition different from identity which sends the set
A to itself. Let T  be the set of semi-transpose tone row, we have
S ∈ T  ⇔ ∃ i = 0, Ti (A) = A
Example 7.4. The tone row of Schönberg’s Ode to Napoleon
(451089et2376) which has the chord diagram D538 is a semi-transpose
tone row because T4 (A) = A.
Definition 7.5. A semi-reverse tone row is a tone row for which
it exists a nonidentity transposition which sends the set −A to the
set A. If R denotes the set of semi-reverse tone row, we have
S ∈ R ⇔ ∃ i = 0, Ti (−A) = A
Example 7.5. The tone row of Luigi Nono’s Canto Sospeso
(9t8e70615243) which has a chord diagram D358 is a semi-reverse
tone row because T7 (−A) = A. The first tone row of Jean
Barraqué’s Concerto (7490612t5e38) which has the chord diagram
D271 is a semi-reverse tone row.
 Serial Knots 185

Many tone rows do not belong to the four previous types. For
instance, the tone rows of Pierre Boulez’s Sonatine for flute and
piano (0e7184392t56), the Cycle de l’artisan furieux of Le Marteau
sans Maı̂tre (314596t78e20), Karel Goeyvaerts’s Opus 2 for 13
instruments (e65478329t10), Jacques Lenot’s Third Sonate for piano
(58726039t14e), Peter Schat’s Polonaise opus 2 (689e2t013457) and
Jan van Vlijmen’s Costruzione per due pianoforti (015473298t6e) do
not belong to the set T ∪ T  ∪ R ∪ R .

7.5 Combinatoriality

The combinatoriality is a notion first introduced by Milton Babbitt.

Since combinatoriality applies to any set, we only consider the combi-
natoriality of hexachords in order to compare the two classifications
of Costère and Babbitt. The basic transformations and their associ-
ated rows are denoted by P (basic row), I (inversion), R (retrograde)
and RI (retrograde of the inversion).
Definition 7.6. A semi-combinatorial tone row is a 12-tone row
such that any of its transformations other than the retrograde can
be transposed so that the first six notes are equivalent to the last six
notes of the original set, not necessarily in the same order.
Example 7.6. The 12-tone row S = 648203951te7 is a semi-
combinatorial tone row because in the first transposition of the inver-
sion T1 I(S) = 795e1t480326, the set of the last six notes is equivalent
to the set of the first six notes of the original set S. Remark that any
set is combinatorial with its own retrograde.
The combinatoriality of rows has been first advised by Ernst
Krenek in 1940. The composer studied the symmetric row
S = 91e2t0846375
composed of the first hexachord A and the transposition of the
inversion
S = A ∪ I11 (A)
Proposition 7.1. The basic transformations denoted by P (basic
row), I (inversion), R (retrograde) and RI (retrograde of the inver-
sion) are the Klein four-groups.
186 A Compendium of Musical Mathematics

The multiplicative table is given by

P I R RI
P P I R RI
I I P RI R
R R RI P I
RI RI R I P
Definition 7.7. The all-combinatorial tone row is a 12-tone row
whose first hexachord forms a 12-tone row with any of its basic trans-
formations (P, I, R, RI) transposed.
Proposition 7.2. The second hexachord of all-combinatorial tone
rows is always a literal transposition (Ti with i = 0) of the first
hexachord.
Example 7.7. The chromatic 12-tone row S = 0123456789te is
formed of the first hexachord A = 012345 whose transposition a
tritone higher leads to the second hexachord:
S = A ∪ T6 (A)
Theorem 7.4. Let A be the first hexachord of a 12-tone row S. The
all-combinatorial rows S are classified in six types of unordered sets
and gathered in four orders:
(1) First order combines A with only one transposition T6
S = A ∪ T6 (A)
A is one of the three unordered sets
A = {0, 1, 2, 3, 4, 5}
A = {0, 2, 3, 4, 5, 7}
A = {0, 2, 4, 5, 7, 9}
(2) Second order combines A with two transpositions T3 and T9
S = A ∪ T3 (A) = A ∪ T9 (A)
A is the unordered set
A = {0, 1, 2, 6, 7, 8}
(3) Third order combines A with three transpositions T2 , T6 , T10
S = A ∪ T2 (A) = A ∪ T6 (A) = A ∪ T10 (A)
 Serial Knots 187

A is the unordered set

A = {0, 1, 4, 5, 8, 9}
(4) Fourth order combines A with six transpositions T1 , T3 , T5 , T7 ,
T9 , T11
S = A ∪ T1 (A) = A ∪ T3 (A) = A ∪ T5 (A)
= A ∪ T7 (A) = A ∪ T9 (A) = A ∪ T11 (A)
A is the unordered set
A = {0, 2, 4, 6, 8, 10}
Theorem 7.5. Any hexachord A and its complement set Ac have
the same interval vector.
Example 7.8. In the 12-tone row of Schönberg’s Fourth String
Quartet S = 0e78312t6549, the two hexachords have the same inter-
val vector
iv(S) = [3, 2, 2, 4, 3, 1]

7.6 Similarity Measures

Tuukka Ilomäki (2008) has studied similarly measures between tone

rows. In order to compare their algebraic and combinatoric proper-
ties, he proposed 17 similarity measures d(X, Y ) between two 12-tone
rows X and Y .
The similarity measure d(X, Y ) between two 12-tone rows X and
Y is a numeric measurement of the similarity of two 12-tone rows.
Since similarity has not a mathematical definition, there are several
similarity measures. In most cases (but not always), the similarity
measure is a metric.
Definition 7.8. A metric d(X, Y ) between two 12-tone rows X and
Y is a function (also called a distance) in the set of nonnegative real
numbers, such that following three axioms are satisfied:
(i) d(X, Y ) = 0 ⇐⇒ X = Y
(ii) d(X, Y ) = d(Y, X)
(iii) d(X, Z) ≤ d(X, Y ) + d(Y, Z)
188 A Compendium of Musical Mathematics

Definition 7.9. The similarity measure is consistent if for any

interval-preserving transformation F (transpositions, inversions,
retrogradations, rotations and their combinations), we have

d(X, Y ) = d(F X, F Y )
Consider the 12-tone row of Jean Barraqué’s Piano Sonata, writ-
ten in compact form X = 46e5102398t7 and the same slightly dis-
turbed tone row Y = 46e8102395t7. The tone row X is also written
in the form of a permutation
0 1 2 3 4 5 6 7 8 9 t e
X=
4 6 e 5 1 0 2 3 9 8 t 7
that we write in cycle notation
X = (0, 4, 1, 6, 2, e, 7, 3, 5 )(6, 9)

and the same for tone row Y

0 1 2 3 4 5 6 7 8 9 t e
Y =
4 6 e 8 1 0 2 3 9 5 t 7
The inverse of Y is
0 1 2 3 4 5 6 7 8 9 t e
Y −1 =
5 4 6 7 0 9 1 e 3 8 t 2
and the product is
0 1 2 3 4 5 6 7 8 9 t e
XY −1 =
0 1 2 9 4 5 6 7 8 3 t e

The cyclic notation of this permutation XY −1 = (3, 9) suggests the

only ranks of the permutation where X and Y differ. In compact
form, the series is written as XY −1 = 0129456783te. Let sn be the
n-th element of the tone row XY −1 = s0 s1 · · · s11 . Thanks to these
definitions, Tuuka Illomäki introduces different measures of similar-
ity. Let’s see the most important. All of these similarity measures are
metrics except the last one: the scattering measure.
Definition 7.10. The derangement between two 12-tone rows X
and Y is the number of points where the n-th elements xn and yn
 Serial Knots 189

are different:

der(X, Y ) = card{n, xn = yn }

it is also the number of points where the n-th element sn of the tone
row XY −1 is different from n:

der(X, Y ) = card{n, sn = n}

Example 7.9. In the previous Barraqué example, der(X, Y ) = 2

because the series differ in two places.
Definition 7.11. The displacement is the distance between the
notes themselves:
11
 11

dep(X, Y ) = |xn − yn | = |sn − n|
n=0 n=0

Example 7.10. In the Barraqué example, dep(X, Y ) = |5 − 8| +

|8 − 5| = 6. This measure is not consistent for circular permutations.
Definition 7.12. The Cayley distance is the number of cycles in
XY −1 .
Example 7.11. In the previous example, the permutation XY −1 =
(3, 9) has a unique cycle.
Definition 7.13. The Ulam distance is 12 reduced by the length of
the greater common segment. The segment common to X and Y is
the set of points where the values are the same.
Example 7.12. In the Barraqué example, the common segment
46e10239t7 has length 10. The Ulam distance between X and Y
is therefore 2.
Definition 7.14. The division is the number of adjacent pairs
(sn , sn+1 ) in XY −1 that are not consecutive ascending pairs
(m, m + 1).
Example 7.13. In the Barraqué example, the tone row
0129456783te has adjacent pairs (0,1), (1,2), (2,9), etc. But only pairs
(2,9), (9,4), (8,3) and (3,t) are not consecutive. The division between
X and Y is thus 4.
190 A Compendium of Musical Mathematics

Proposition 7.3. The division is a consistent metric.

Definition 7.15. The scattering measures the difference between
the elements of XY −1 :
10

diff(X, Y ) = |sn+1 − sn | − 11
n=0
Remark 7.1. This measure was set up by Morris (1987). It is not a
metric.
Other interesting measures like nesting are to build a distance
between all common subsegments of two measures. We refer to the
work of Tuuka Illomäki for more details.

7.7 Serial Groups

In this section, we consider permutations associated with chord dia-

grams in order to measure the generating power of these diagrams.
Each diagram represents a set of tone rows. But the symmetries limit
the number of tone rows associated with a chord diagram. We define
a group whose order represents this generating power.
Each chord diagram is associated with a permutation in the fol-
lowing way. For each chord diagram, we number the vertices of the
diagram clockwise and consider the pairs of numbers corresponding
to linked vertices. The set of the pairs defines the permutation of the
chord diagram.
Example 7.14. The diagram D358 is the diagram of the tone row
of Anton Webern’s Symphony opus 21, Luigi Nono’s Canto Sospeso,
Alban Berg’s Suite lyrique and Karlheinz Stockhausen’s Klavierstück
IX. The labeling of the vertices yields to the pairs
(0, 1)(2, 11)(3, 10)(4, 9)(5, 8)(6, 7)
These pairs define the permutation P358 associated with the diagram
D358 :
0 1 2 3 4 5 6 7 8 9 10 11
P358 =
1 0 11 10 9 8 7 6 5 4 3 2
 Serial Knots 191

We set the following definition.

Definition 7.16. Let Dn be a chord diagram. We denote Pn the
associated permutation, T1 the transposition of a semitone up and
I the inversion. The group of the chord diagram Dn is the group
generated by the three operators:
Gn = Pn , T1 , I
The order of this group is called the generating power of the dia-
gram Dn .
Example 7.15. The group associated with the diagram D358 is the
group
G358 = P358 , T1 , I
The GAP software gives the presentation of this group in terms of
generators and relations:

G358 = a, b, c | a2 = b12 = c2 = ab−1 c = 1
The group has |G358 | = 24 elements. This group and the dihedral
group have the same number of elements.
Example 7.16. The 12-tone row (926541380et7) of diagram D554
is used by B.A. Zimmermann in his opera Die Soldaten and also by
Pierre Boulez in the Cycle des bourreaux de solitude (Piece 6) of Le
Marteau sans Maı̂tre.(23456789te01). The associated group

G554 = a, b, c | a2 = c2 = abab−1 = (ac)2 = (b−1 c)2 = b6 a = 1
has 24 elements.
Example 7.17. The 12-tone row (et2376845019) used by Anton
Webern in his Concerto for nine instruments, opus 24 is associated
with the diagram D549 whose group
 
a, b, c | a2 = c2 = (b−1 c)2 = (acb)2
G549 =
= (ab2 c)2 = (ab−1 )4 = (abab−1 )3 = 1
is of order 192.
192 A Compendium of Musical Mathematics

Example 7.18. The 12-tone row (5t697821304e) used in

Schönberg’s Ode to Napoleon and Kees van Baaren’s Variations in
Isometrical Series is associated with the diagram D538 . The associ-
ated group generated by the three operators G538 = P538 , T1 , I has
7680 elements.

Example 7.19. The 12-tone row (te36828715904) used by Dal-

lapiccola in Quaderno Musicale di Annalibera is associated with
the diagram D56 . The group generated by the three operators
G56 = P56 , T1 , I has the presentation
 
a, b, c | a2 = c2 = (b−1 c)2 = (ac)4 = acb−1 (abc)2
G56 =
= (acb3 )2 = (ab2 ab−2 )2 = b12 = (acab−2 )4

and order 28800.

Example 7.20. Many diagrams have a maximum of generating

power. It is the case, for instance, of the tone row D244 used in
Schönberg’s Variations for Orchestra, of the tone row of diagram
D16 used in Jacques Lenot’s Third Sonate for piano and of the tone
row of diagram D353 used in Milton Babbitt’s Three Compositions
for piano. In each case, the associated group is isomorphic to the
permutation group S12 of order 479,001,600.

The analytical study of the groups associated with all the string
diagrams leads to the following result.

Theorem 7.6. The 554 chord diagrams generate 27 serial groups

which form a subset of the 301 transitive permutation groups Gr
(where r is the group index in the GAP software). Among these 27
groups, only two are primitive: G218 of order 1320 which is the group
P GL(2, 11) and G301 which is the permutation group S12 . 431 chord
diagrams are associated with the permutation group S12 .

The lattice of serial groups is represented in Fig. 7.2.

Definition 7.17. Let S be a 12-tone row with a chord diagram D

and G be the serial group of D. The action of G on S is said to be
globally stable if for any g ∈ G, the series g(S) has a chord diagram
associated with G or to one of its subgroups.
 Serial Knots 193

Fig. 7.2. The serial groups Gr = Pn , T, I.

Example 7.21. (1) The chord diagram of the 12-tone row S =

061724e3t895 is D50 of group G118 of order 216. For all g ∈ G118 ,
the tone row g(S) has a chord diagram D50 or D358 . But D358 has
a group G12 of order 24 which is a subgroup of G118 . (2) The chord
diagram of the 12-tone row S = 08673e9241t5 is D474 and its group
G118 . For all g ∈ G118 , the tone row g(S) has a chord diagram D359
or D474 .
Theorem 7.7. Only nine chord diagrams are stable under the action
of their serial group. Let S be a 12-tone row of chord diagram D and
G be the serial group of D. For all g ∈ G, the 12-tone row g(S) has
the same chord diagram D. These are the diagrams : D358 and D554
of serial group G12 of order 24, D444 and D509 of serial group G28 of
order 48, D491 of serial group G86 of order 192, D487 of serial group
G125 of order 288, D497 et D508 of serial group G152 of order 384 and
D530 of serial group G186 of order 768.

7.8 Exercises

Exercise 7.1
Build all chord diagrams in Z6 . List their Gaussian words and their
intervallic structure. Show that the symmetry of the diagrams is
such that there is no difference between the action of the cyclic
group and the action of the dihedral group. Find this result using
194 A Compendium of Musical Mathematics

Khruzin’s formula. By removing the notes which do not belong to

Z6 in the series 451089et2376, give a series of six notes. Deduce from
this series (by changing the order of the notes) an all-interval series.
Verify that its chord diagram is the only one capable of encoding
all-interval series.
Exercise 7.2
Let us consider the permutation

a = (0, 1)(2, 11)(3, 10)(4, 9)(5, 8)(6, 7)

and the group


G = a, b, c | a2 = b12 = c2 = ab−1 c = 1

where b is the translation T1 (x) = x + 1 and c is the inversion

c(x) = −x mod 12. Write down the 24 elements of G classified by
powers of b. Draw the chord diagram of the series used in Schönberg’s
Ode to Napoleon 5e697821304t and show that the action of the per-
mutation a on the series does not change the chord diagram.
Exercise 7.3
Let us consider the series of Milton Babbitt’s Second string quartet
03e4128t5976. Study the properties of this series.
Exercise 7.4
Let us consider the series of Bruno Maderna’s Concerto for piano,
S = 2e61094583t7. Study the properties of this series. Show that
S∈/ T ∪ R ∪ T  ∪ R .
Exercise 7.5
Let us consider the series of Jean Barraqué’s Au-delà du hasard,
S = 087142A3B569 and the permutation

0123456789te
σ=
97823e t10645
= (096t4328)(17)(5e)

Prove that σ 8 (S) = S and draw the chord diagrams for each
σ k (S),for k = 1, . . . , 8. Draw the graph of the permutational evolu-
tion of the series used by Barraqué.
 Serial Knots 195

Exercise 7.6
Let us consider two series of Elisabeth Lutyens’s Requiem: serie
A = 32e78014569t and serie B = 370e8465t912. Compute the Cayley
distance of the two series.

References

Costère, E. (1962). Mort ou transfiguration de l’harmonie, Presses Universitaires

de France, Paris.
Eimert, H. (1963). Grundlagen der musikalischen Reihentechnik, Universal
Edition, Wien.
Fripertinger, H. and Lackner, P. (2015). Tone rows and tropes, Journal of
Mathematics and Music 9(2), 111–172.
Herzog, E. (1964). Úplný rejstřı́k dvanáctitónových všeintervalových řad [A com-
plete index of the all-intervall series], Nové cesty hudby: sbornı́k studiı́
o novodobých skladebných směrech a vědeckých pohledech na hudbu [New
Paths to Music: An Anthology of Studies on Modern Composition Trends
and Scientific Perspectives on Music], Státnı́ hudebnı́ vydavatelstvı́, Praha.
Jedrzejewski, F. (2021). La musique dodécaphonique et sérielle: une nouvelle
histoire, Brepols, Turnhout.
Khruzin, A. (2000). Enumeration of chord diagrams. arXiv:math/0008209.
Ilomäki, T. (2008). On the Similarity of Twelve-Tone Rows, Sibelius Academy,
Helsinki.
Morris, R. (1977). On the generation of multiple order-function twelve-tone rows,
Journal of Music Theory 21(2), 238–263.
Morris, R. (1985). Set-type saturation among twelve-tone rows, Perspectives of
New Music 22(1–2), 187–217.
Morris, R. (1988). Generalizing rotational arrays, Journal of Music Theory 32(1),
75–132.
Morris, R. (1989). Some compositional and analytic applications of T-matrices,
Integral 3, 37–66.
Morris, R. (2007). Mathematics and the twelve-tone system: Past, present, and
future, Perspectives of New Music 45(2), 76–107.
Parzysz, B. (1984). Musique et mathématique, 53, Publication de l’APMEP
(Association des professeurs de mathématiques de l’enseignement public),
Paris.
Reiner, D. (1985). Enumeration in music theory, American Mathematical Monthly
92, 51–54.
Riotte, A. (1963). Génération des cycles équilibrés, Rapport interne 353, Ispra,
Euratom.
Riotte, A. and Mesnage, M. (2006). Formalismes et modèles musicaux, 2 Vols.,
Ircam-Delatour, Sampzon.
This page intentionally left blank
 Chapter 8

Combinatorial Designs

Combinatorial designs are blocks of numbers of the same size that

have special properties. They have been used by Carlton Gamer
in connection with projective geometry in musical pieces built on
tempered N -tone systems (Gamer and Wilson, 2003) and much more
systematically by Tom Johnson (Andreatta et al., 2009; Jedrzejewski
and Johnson, 2014; Johnson, 2019; Delor, 2021) who was inspired by
the work of two mathematicians Charles J. Colbourn and Jeffrey H.
Dinitz (2006).

8.1 Difference Sets

Definition 8.1. Let G be an Abelian group of order v. A (v, k, λ)

difference set is a k-subset D ⊆ G such that every nonzero element
of G can be expressed as a difference x − y with x, y ∈ D in exactly
λ ways. In other words, such that every nonzero element of G appears
the same number of times in the multiset

Δ = {d2 − d1 | d1 , d2 ∈ D, d1 = d2 }

n = k − λ is the order of the difference set.

Theorem 8.1. If D is a (v, k, λ) difference set, then

k(k − 1) = v(v − 1)

197
198 A Compendium of Musical Mathematics

Example 8.1. D = {1, 2, 4} is a (7,3,1) difference set. D = {0, 1, 2, 4,

5, 8, 10} is a difference set of Z11 . The set D = {a, a2 , a4 , b, b2 } is a
difference set of the group
 
G = a, b | a7 = b3 = 1, ab = ba

Definition 8.2. Two difference sets D1 and D2 are equivalent in

a group G if for some g ∈ G and some automorphism ϕ of G,
D2 = g + ϕ(D1 ).
Proposition 8.1. The complement of a (v, k, λ) difference set in G
is a (v, v − k, λ + v − 2k) difference set.
Definition 8.3. A (v, k, λ) tiling of G is a collection {D1 , D2 , . . . ,
Dr } of mutually disjoint (v, k, λ) difference sets such that the union
of Dj is G\{0}:

D1 ∪ D2 ∪ · · · ∪ Dr = G\{0}

Example 8.2. This example may be of interest to composers

who work in 31-tone universe. The sets D1 = {1, 5, 11, 24, 25, 27},
D2 = {2, 10, 17, 19, 22, 23}, D3 = {3, 4, 7, 13, 15, 20}, D4 = {6, 8, 9,
14, 26, 30} and D5 = {12, 16, 18, 21, 28, 29} are a tiling of (31, 6, 1)
of Z31 \{0}.
Definition 8.4. Let D be a (v, k, λ) difference set in G. An integer
p relatively prime to the order of G is a multiplier for D if Mp (D) =
D + g for some g ∈ G, where Mp is the multiplication by p defined by

Mp (x) = px mod v

Example 8.3. Consider D = {2, 3, 5, 11} in Z13 . The integer p = 3

is a multiplier because

M3 (D) = {6, 9, 2, 7} = D + 4

Theorem 8.2 (First multiplier theorem). Let D be a (v, k, λ)

difference set and p be a prime dividing n = k − λ and such that
gcd(p, v) = 1. If p > λ, then p is a multiplier of D.
Example 8.4. The set D = {1, 3, 13, 16, 17} is a (21, 5, 1) difference
set. Then p = 2 is a multiplier.
 Combinatorial Designs 199

Theorem 8.3 (Second multiplier theorem). Let D be a (v, k, λ)

difference set and assume that n = k − λ is a power of a prime p
with gcd(p, v) = 1. Then p is a multiplier of D.

Example 8.5. The set D = {0, 1, 2, 4, 5, 8, 10} of Z15 is a (15, 7, 3)

difference set. Since n = 4 = 22 , p = 2 is a multiplier.

8.2 Block Designs

A combinatorial design is a (k, b)-matrix of positive integers such that

each column is called a block, satisfying nice substructure properties.
Composer Tom Johnson has used combinatorial designs in music
since 2003.

Definition 8.5. A t-design t-(v, k, λ) is a pair D = (X, B) where X

is a v-set (X = Zv ) and B a collection of k-subsets of X called blocks
such that every t-subset of X is contained in exactly λ blocks. D is
simple if it has no repeated block.

Block designs have different names depending upon their param-

eter values: 2-(v, k, λ) are called Balanced Incomplete Block Design
(BIBD) and simply denoted by (v, k, λ), for λ = 1, t-(v, k, 1)
are Steiner Systems, t-(v, 3, 1) are Triple Systems (TS), 2-(v, 3, 1)
are Steiner Triple Systems (STS) and 2-(v, 4, 1) are called Steiner
Quadruple Systems (SQS). There are no known examples of nontriv-
ial t-designs with t ≥ 6. For t = 5, the design 5-(24, 8, 1) is a well-
known Steiner system. A symmetric design is a BIBD (v, k, λ) with
b = v. Of course, block designs are considered up to isomorphism.

Definition 8.6. Two t-designs (X1 , B1 ) and (X2 , B2 ) are isomorphic

if there is a bijection ϕ : X1 → X2 such that ϕ(B1 ) = B2 .

Example 8.6. The simple example is the Fano plane or design

(7, 3, 1). The design includes 7 blocks of 3 digits, taken through the
set of the first numbers X = {0, 1, . . . , 6}. Each pair of numbers
appears in only one block: {0, 1} appears in the first block, {0, 2}
appears in the second block, {0, 3} appears in the first block and so
on, until {5, 6} which appears in the fifth block. Each element of X
200 A Compendium of Musical Mathematics

appears three times in three different blocks. These properties guar-

antee the variety and the interest that one can have by transposing
these numbers to musical objects:

0 0 0 1 1 2 3
1 2 4 2 5 3 4
3 6 5 4 6 5 6

Example 8.7. This design is isomorphic to the following design

using the permutation in cycle notation (2, 3)(4, 5, 6):

0 0 0 1 1 2 2
1 3 5 3 4 3 4
2 4 6 5 6 6 5

The dual of the Fano plane exchanges points and lines (see
Fig. 8.1).

Remark 8.1. It is sometimes difficult to know how many nonisomor-

phic designs exist. For instance, there exists a unique (6, 3, 2) design,
a unique (9, 3, 1), but four nonisomorphic (8, 4, 3) designs, 11 noni-
somorphic (9, 4, 3) designs, three nonisomorphic (10, 4, 2) designs, a
unique (11, 5, 2), a unique (13, 4, 1) design, etc. There are 242,995,846
nonisomorphic (12, 4, 3) designs, and 11,084,874,829 nonisomorphic
(19, 3, 1) designs. This opens up great prospects for composers who
would like to associate the v-set with the musical notes. A long list
can be found in the Handbook of Combinatorial Designs of Colbourn

Fig. 8.1. The Fano plane and its dual.

 Combinatorial Designs 201

and Dinitz (2006). It is not known how many nonisomorphic (12, 4, 3)

designs exist.
Proposition 8.2. The number of blocks of a t-(v, k, λ) design is
given by the formula
v! (k − t)!
b=λ
(v − t)! k!
and the number of blocks incident with any point is
(v − 1)! (k − t)!
r=λ
(v − t)! (k − 1)!
This leads to the simple relation
bk = vr
Important properties of a design can be determined by the prop-
erties of its incidence matrix.
Definition 8.7. The incidence matrix of the (v, k, λ) design is the
matrix N = (nij ) of v rows and b columns such that the element nij
is 1 if the element i is in the block j and 0 otherwise.
Theorem 8.4 (Fisher’s inequality). A 2-(v, k, λ) design with
v > k has a number of blocks b ≥ v.
Proof. Let N be an incidence matrix of a (v, k, λ) design. Then
N satisfies
N N t = (r − λ)I + λJ
where I denotes the v × v identity matrix and J denotes a v × v
matrix with all entries equal to 1. Using the fact that for a n × n
matrix
det(aI + bJ) = (a + nb)an−1
we get
det(N N t ) = rk(r − λ)v−1 = 0
because v > k implies r > λ. This show that (N N t ) has full rank.
Since N and N N t have the same rank, the rank of N is v. But the
rank cannot exceed the number of columns, thus b ≥ v. 
202 A Compendium of Musical Mathematics

Definition 8.8. The complement of t-(v, k, λ) design D = (X, B) is

the design D c = (X, X\B) of parameters t − (v, v − k, μ) with
   
v−t  v−t (v − k)! (k − t)!
μ=λ =λ
k k−t (v − t − k)! k!

D and D c have the same number of blocks.

Proposition 8.3. For t = 2, the block design D with b blocks has a

complement D c with b blocks and parameters (v, v − k, b − 2r + λ).

Example 8.8. For instance, the complement of (7, 3, 1) is the design

(7, 4, 2) with blocks {0, 1, 2}c = {3, 4, 5, 6}, etc.

Definition 8.9. A (v, k, λ) design is symmetric if v = b.

Theorem 8.5 (Bruck–Ryser–Chowla, 1950). Let v, k, λ be in-

tegers satisfying λ(v − 1) = k(k − 1) for which a symmetric 2-(v, k, λ)
design exists. Then we have the following:

(1) If v is even, then (k − λ) is a perfect square.

(2) If v is odd, the diophantine equation

z 2 = (k − λ)x2 + (−1)(v−1)/2 λy 2

has a nonzero solution in integers x, y, z.

Proof. The proof of (1) is straightforward. Since

det(N N t ) = rk(r − λ)v−1 = 0

and k = r,

det(N ) = k2 (k − λ)v−1

Since det(N ) is an integer and v − 1 is odd, then (k − λ) is a perfect

square. We leave the proof of the second part which is too technical
for our purpose. 

Example 8.9. The symmetric design (49, 16, 5) leads to the dio-
phantine equation z 2 = 11x2 + 5y 2 which (1, 1, 4) is a nonzero
solution.
 Combinatorial Designs 203

Remark 8.2. It has been proved by Lam in 1989 that the converse
of the Bruck–Ryser–Chowla (BRC) theorem is false. The projec-
tive planes are symmetric designs with parameters (v, k, λ) =
(n2 + n + 1, n + 1, 1), where n is the order of the projective plane.
Lam shows that there are no projective plane of order 10. If there
were a projective plane of order n = 10, it would be a symmetric
design (111, 11, 1). Since 111 is odd, and the diophantine equation
z 2 = 10x2 − y 2 has the solution (1, 3, 1), the design (111, 11, 1) passes
the BRC test. But since there are no projective plane of order 10,
the BRC conditions are not sufficient to guarantee the existence of
symmetric designs.

Proposition 8.4. If a symmetric design (v, k, λ) exists with v = 4m,

then (v, k, λ) = (4m2 , 2m2 − m, m2 − m) for some integer m.

When v is different from 12, the problem arises as to how to musi-

cally use the design block, assuming that the v-set is a set of pitches.
That’s a question Tom Johnson asked himself while composing Net-
works. This work is based on design (6, 3, 2). The composer wrote
the following:
“For (6,3,2), one needs a six-note scale, but which six notes?
The combinatorial logic does not imply that note 1 is higher
or lower than note 6, or that there is any particular proximity
between notes 3 and 4, though the musical goal was pretty clear
as far as I was concerned. A particular group of chords should
sound homogeneous, belonging to the same musical world, but if
the scales are always constructed in one standard way, the net-
work all sound about the same, which they shouldn’t. Above
all, it was important to stop hearing the chords as tonal and
atonal. The music I was finding here was strictly combinato-
rial, which is a universal concept that includes both tonal and
atonal music, and it was important to bring the two together,
despite the fact that the history of music in the last century
has generally viewed the two traditions as totally opposed. Of
course, there was no reason to limit myself to the chromatic
scale either, but I feel more comfortable with this and prefer to
leave microtonal experiments to others”

(Johnson, 2019: p. 482).

As can be seen in Fig. 8.2, each triangle is a block of the design

(6, 3, 2) which has 10 blocks (in white in the figure). The hatched
triangles are a duplication of the same blocks. Thus, the composition
204 A Compendium of Musical Mathematics

Fig. 8.2. Tom Johnson, Networks, design (6, 3, 2).

is based on 20 chords or blocks. Tom Johnson gave each number

a musical equivalent: 1 (D), 2 (F), 3 (F sharp), 4 (A), 5 (B flat),
6 (D flat).

8.3 Resolvable Designs

An important class of designs is the class of resolvable designs.

Definition 8.10. A parallel class in a design is a set of blocks that
partition the point set.
Definition 8.11. A design (v, k, λ) is resolvable if its blocks can be
partitioned into parallel classes.
Example 8.10. For instance, (9, 3, 1) is resolvable as shown in the
following table. The 12 blocks of this design can be arranged by
groups of 3 blocks such that the union of these blocks (a column in
the following table) is the point set.
 Combinatorial Designs 205

(0,1,2) (0,3,6) (0,4,8) (0,5,7)

(3,4,5) (1,4,7) (1,5,6) (1,3,8)
(6,7,8) (2,5,8) (2,3,7) (2,4,6)
Theorem 8.6. For every positive n, there exists a resolvable
(2n, n, 1) design.
Example 8.11. Twelve (2008), 12 short pieces for piano, composed
by Tom Johnson is based on the design (12, 4, 3). This design has 33
blocks, and each block has 4 elements. On the graph drawn by the
composer (see Fig. 8.3), we see the 33 blocks connected to each other
when they have no elements in common. So the hatched triangles
represent the parallel classes.

Fig. 8.3. Tom Johnson, Twelve, design (12,4,3).

206 A Compendium of Musical Mathematics

8.4 Kirkman’s Ladies

In 1850, Reverend Thomas Penyngton Kirkman (1806–1895) posed

the so-called schoolgirls problem “Fifteen young ladies in a school
walk out abreast for seven days in succession: it is required to arrange
them daily, so that no two walk twice abreast.” Kirkman’s ladies
can be considered the beginning of block designs distributed into 35
blocks of 3 elements, with each pairs coming together exactly once. It
has been shown that this problem leads to a resolvable Steiner Triple
system 2-(v, 4, 1) if and only if v ≡ 3 (mod 6). The Kirkman problem
is solved for v = 15. There are seven nonisomorphic solutions. One
solution is given in the following table.

Monday (0,1,2) (3,9,11) (4,7,13) (5,8,14) (6,10,12)

Tuesday (0,3,4) (1,8,10) (2,10,14) (5,7,11) (6,9,13)
Wednesday (0,5,6) (1,7,9) (2,11,13) (3,12,14) (4,8,10)
Thursday (1,3,5) (0,10,13) (2,7,12) (4,9,14) (6,8,11)
Friday (1,4,6) (0,11,14) (2,8,9) (3,7,10) (5,12,13)
Saturday (2,3,6) (0,7,8) (1,13,14) (4,11,12) (5,9,10)
Sunday (2,4,5) (0,9,12) (1,10,11) (3,8,13) (6,7,14)

Composer Tom Johnson has used this design in a piece called

Kirkman’s ladies. If one wonders if it is possible for the ladies to con-
tinue their walks for a complete semester of 13 weeks, so as to include
all 15-choose-3 = 455 possible blocks leads to a huge structure called
a (15, 3, 1) large designs. This structure was used by Tom Johnson
in his work. The composer has taken care to represent in Fig. 8.4
parallel classes linked to the Fano plane.

8.5 Block Design Drawings

In the representations of block designs, the mathematician is inter-

ested in relations resulting from the action of a symmetry group.
Most of the time, it is easy to demonstrate a generation of designs
by the action of translations on a set of given blocks. But that does
not allow drawing all combinatorial designs. For the music composer,
 Combinatorial Designs 207

Fig. 8.4. Tom Johnson, Kirkman’s ladies, design (15,3,1)

this is not enough and he will often consider other relations, such as
the absence of common points or the movement of only one point of
each block.
The easy case is when blocks are constructed from a set of gen-
erators B = B | T1v (B) ≡ 1 with action of the cyclic group. These
designs are represented by cyclic representations. A special case are
the designs
 m 
p − 1 pm−1 − 1 pm−1 − 1
, ,
p−1 p−1 p−1
208 A Compendium of Musical Mathematics

of the projective geometry P G(m − 1, p) depicted in the following

table. The last column indicates the generator block of the design:

(7,3,1) PG(2,2) (0,1,3)

(13,4,1) PG(2,3) (0,1,3,9)
(21,5,1) PG(2,4) (0,1,4,14,16)
(31,6,1) PG(2,5) (0,1,3,8,12,18)
(57,8,1) PG(2,7) (0,1,3,13,32,36,43,52)
(73,9,1) PG(2,8) (0,1,3,7,15,31,36,54,63)
(91,10,1) PG(2,9) (0,1,3,9,27,49,56,61,77,81)

Another case is given by Netto’s theorem (1893).

Theorem 8.7 (Netto, 1893). Let p be a prime, n ≥ 1, pn ≡ 1

(mod 6). Let Fpn be a finite field on X of size pn = 6t + 1 with

Fig. 8.5. Design (9, 4, 3) showing blocks as connected pairs.

 Combinatorial Designs 209

0 as its zero element and α a primitive root of unity. The sets

Bi = {αi , αi+2t , αi+4t } mod pn for i = 1, 2, . . . , t − 1 are genera-
tors (Tj (B) = j + B mod pn ) of the set blocks of a Steiner triple
system on X.

Sometimes it’s easier to break the design down into subdesigns or

to inflate it. For instance, the (9, 4, 3) design of 18 blocks is inflated
into (9, 2, 3) designs of 36 blocks. In this case, the representation of
the design (9, 4, 3) is based on connected pairs of (9, 3, 2) as I have
done it (see Fig. 8.5).

8.6 Tom Johnson’s Graphs

Tom Johnson is the composer who has worked the most on block
designs. In 2009, he composed Septet II for two flutes, oboe, clarinet,
two violins and viola based on the Fano plane (7, 3, 1). In Vermont
Rhythms (2008), a 42 × 11 rhythm, he used the design (11, 6, 3) of
11 blocks.
Fifty-Five Chords (2008) for organ, 23 minutes of organ music
all derived from a (11, 4, 6) block design of 55 blocks. This last
piece raises the question of the graphical representation of a block
design. Johnson gives several graphs of the same design: Cosmo-
logical view in which every single chord has no notes in common
with exactly four chords; Pentagonal view where each chord has one
pair of notes in common with one chord, the other pair in common
with one other chord, and no notes in common with the adjacent
chords; Spider web view linking chords with three notes in com-
mon; and Starfish view where three pairs of notes combine to form
three chords (see Jedrzejewski and Johnson, 2014). On the wheel of
the graph drawn by the composer (see Fig. 8.6), the first block is
on the left opening (11, 10, 3, 2). Blocks are connected if they have
three notes in common, and move clockwise until one reach the
block (2, 9, 3, 10), which has also three notes in common with the
first one.
Block Design for piano (2005) is based on design 4-(12, 6, 10) of
330 blocks. It could be built using 30 base blocks and one automor-
phism that adds 1 to each point except 11 which remains always
the same. In the commentary he gives of his work, the composer
shows how the block design technique is for him as important as the
210 A Compendium of Musical Mathematics

Fig. 8.6. Tom Johnson, Fifty-Five Chords, design (11, 4, 6).

discovery of the 12-tone rows by Schoenberg and responds to the

same ideal of composition. He wrote the following:
The configuration here is technically a 4-(12, 6, 10) block design,
which in my music terms means that there are twelve notes,
distributed into six-note arpeggios, in such a way that every
combination of four particular notes comes together exactly ten
times in ten different arpeggios. In fact, it is also true that every
combination of three notes comes together thirty times in thirty
different arpeggios, every pair of notes occurs seventy-five times
in seventy-five different arpeggios, and each of the twelve notes
occurs one hundred and sixty-five times in exactly half of the
three hundred and thirty arpeggios. In a way, the piece is a
realization of Schönberg’s ideal, with exactly equal emphasis of
each twelve notes
(Johnson, 2019: p. 470).
 Combinatorial Designs 211

8.7 Exercises

Exercise 8.1
Verify that D = {1, a, a3 , b, a2 b2 } is a difference set in
 
G = a, b | a7 = b3 = 1, ba = a2 b
Exercise 8.2
Verify that D = {a, a2 , a4 , b, b2 } is a difference set in
 
G = a, b | a7 = b3 = 1, ba = ab
Exercise 8.3
Verify that D = {1, 3, 13, 16, 17} is a (additive) difference set (21,5,1).
Find a multiplier.
Exercise 8.4
In the additive group Z21 , find a multiplier of the difference set
D = {1, 4, 5, 10, 12}.
Exercise 8.5
Let G = Z31 . Find a multiplier of the difference set D = {1, 5, 11,
24, 25, 27}.
Exercise 8.6
Prove that it cannot exist a symmetric design (67, 12, 2).
Exercise 8.7
Consider the block design (7,3,2) of 14 blocks:
1 1 1 1 1 1 2 2 2 2 3 3 3 3
2 2 3 4 5 6 3 4 5 6 4 4 5 5
3 4 5 6 7 7 7 5 6 7 6 7 6 7
Find a graphic representation.

References

Andreatta, M., Jedrzejewski, F. and Johnson, T. (2009). Musical experiences with

block designs. In: Chew, E., Childs, A. and Chuan, C.-H. (dir.) Mathematics
and Computation in Music, Springer, Berlin, 154–165.
212 A Compendium of Musical Mathematics

Colbourn, C. and Dinitz, J. (2006). Handbook of Combinatorial Designs, 2nd edn.,

Chapman and Hall, Boca Raton.
Delor, G. (2021). Tom Johnson ou la musique logique, L’Harmattan, Paris.
Gamer, C. and Wilson, R. (2003). Microtones and projective planes. In: Music
and Mathematics. From Pythagoras to Fractals, Oxford University Press,
Oxford, 146–161.
Jedrzejewski, F. and Johnson, T. (2014). Looking at Numbers, Birkhauser, Bâle
(Switzerland).
Johnson, T. (2019). Finding Music. Writings 1961–2018, MusikTexte, Köln.
 Chapter 9

Tuning Systems

Although most music is composed in the 12-tone equal system, it has

not always been so in Western music. Many tunings were invented
before the establishment of the equal temperament. In addition, from
the beginning of the 20th century, attention has been paid to the
systems obtained by dividing the octave (or another periodic interval)
into other numbers of divisions (see Jedrzejewski, 2014a). Historical
temperaments are described in many books, for instance, (Barbour,
1951; Jedrzejewski, 2002). The reader may also refer to the writings
of Christiaan Huygens and Joseph Sauveur, recently published, see
Caddeo et al. (2021) and Jedrzejewski (2021).

9.1 Cents and Beats

The science of sound studies the communication of sound messages

created by vibrations in the propagation medium. Sounds are created
by compression and expansion of gas molecules in the air. The fre-
quency of the wave is the number of complete oscillations that occur
in one second and expressed in Hertz. A sound is usually described by
its wave nature: these are vibrations transmitted through a medium
with frequencies in the approximate range of 16–20,000 Hz. The stan-
dard pitch or Concert A is defined as 440 Hz. By Fourier theorem,
complex sound waves can be decomposed into a family of simple sine
waves, each of which is characterized by its frequency, amplitude
and phase. These components are called partials (or overtones) and

213
214 A Compendium of Musical Mathematics

the collection of partials is called the spectrum of the sound. This

spectrum consists of a collection of harmonic partials with frequen-
cies f, 2f, 3f, . . . plus some other unrelated inharmonic partials or
noises. Musical intervals are defined by ratios of frequencies and not
directly by the frequencies themselves. The Fechner law states that
the perceived pitch is proportional to the logarithm of the frequency.
The ratios of the frequency are measured in cents. The musical inter-
val of two sounds of frequencies f1 and f2 is given by
1200 log 2 (f2 /f1 ) cents
In classical music and Western music in general, the most common
tuning system is the 12-tone equal temperament or 12 equal division
of the octave (12-EDO), which divides the octave into 12 equal parts.
The smallest interval is called a semitone or half step. Its ratio is 21/2
(100 cents). There are 1200 cents in the octave of ratio 2/1. In the
N -EDO, the octave is divided into N tones. Two consecutive tones
are separated by the interval of 21/N or 1200/N cents.
The savart is also a unit of measurement for pitch intervals. The
musical interval of two frequencies f1 and f2 is given by
1000 log 10 (f2 /f1 ) savarts
Félix Savart has divided the octave on 43 parts called merides, and
each meride is 7 heptamerides. The octave is thus divided in 301
heptamerides which is practically the same as a savart (see Caddeo
et al., 2021 for details).
The usual ratios of the just intonation scale are unison (1), minor
tone (10/9), major tone (9/8), minor third (6/5), major third (5/4),
fourth (4/3) and fifth (3/2).
The sum of two sine waves of closed but distinct frequencies is
perceived as a single sound whose intensity slowly oscillates from
large to small values. These vibrations are called beats. When the
amplitude is large, the interference is said to be constructive and
when the amplitude shrinks to zero, the interference is destructive.
If the frequencies differ by an amount Δf , the resulting sound
sin(2πf t) + sin(2π(f + Δf )t) = 2 cos(πΔf t) sin(2π(f + Δf /2)t)
can be viewed as a sine wave of frequency f + Δf /2 with a slowly
varying envelope A = 2 cos(πΔf t) of frequency Δf . The number
 Tuning Systems 215

of beats per second is the difference of the two frequencies. The

beating can be viewed as alternating regions of constructive and
destructive interference. Two sounds of 440 Hz and 444 Hz cre-
ate four beats per second. Beats are also considered between par-
tials. The third partial of A 440 Hz and the second partial of
E 659 Hz create

3f1 − 2f2 = 3 × 440 − 2 × 659 = 2 beats/second

A tuning system is a choice of number and spacing of frequency values

used.

9.2 Some Commas

In music, a comma is a very small interval, usually less than one

half of the tone. It is defined as the interval between two sequences
of pure intervals. The Pythagorean comma is the difference between
twelve just fifths and seven octaves:

(3/2)12 312 531441

P = 7
= 19
= ≈ 23 cents
2 2 524288
The syntonic comma is the difference between a major Pythagorean
third (81/64) and a just third (5/4):

81/64 34 81
K= = 4 = ≈ 22 cents
5/4 2 ·5 80

The syntonic comma is very close in value to the Pythagorean

comma. The difference is called the schisma. This represents a
frequency

P 531441/524288 32805 38 · 5
S= = = = 15 ≈ 2 cents
K 81/80 32768 2

The Holderian comma is the semitone in the 53-EDO:

CH = 21/53 ≈ 23 cents
216 A Compendium of Musical Mathematics

The Fokker diesis is the 1/31 part of one octave:

CF = 21/31 ≈ 39 cents

The septimal comma is the difference between the Pythagorean

minor seventh (16/9) and the just minor seventh (7/4):

16/9 26 64
C7 = = 2 = ≈ 27 cents
7/4 3 ·7 63

The leimma is the difference between three octaves and five just
fifths:

23 28 256
L= 5
= 5
= ≈ 90 cents
(3/2) 3 243

The apotome is the difference between seven just fifths and four
octaves:

(3/2)7 37 2187
A= = = ≈ 114 cents
24 211 2048

9.3 Historical Temperaments

As we cannot create a cycle of pure fifths (3/2) that repeats itself

exactly after 12 steps, there have been a variety of different systems
of temperament.
In a 12-tone temperament, every fifth can be tempered by a frac-
tion ai of syntonic comma and an amount bi of the schisma. If ai = bi ,
then the fifths are tempered by a fraction ai of the Pythagorean
comma. The difference between the fifth of the temperament and
the pure fifth (3/2) is denoted by

ΔFi = ai K + bi S

In the same way, the difference between the major third of the tem-
perament and the pure major third (5/4) is denoted by ΔTM and
 Tuning Systems 217

the difference between the minor third of the temperament and the
pure minor third (6/5) is denoted by ΔTm :

N. Fréq. ΔF ΔTM ΔTm

0 1 a 1 K + b1 S (1 + s4 )K + r4 S s9 K + (1 + r9 )S
1 37 2−11 ks7 sr7 a8 K + b8 S (1 + s11
8 )K + r8 S
11
−(1 + s75 )K − r57 S
2 32 2−3 ks2 sr2 a 3 K + b3 S (1 + s63 )K + r36 S s11 11
3 K + (1 + r3 )S
9 −14 s9 r9
3 3 2 k s a10 K + b10 S −s2 K − (1 + r2 )S −(1 + s7 )K − r79 S
9 9 9

4 −6 s4 r4
4 3 2 k s a 5 K + b5 S (1 + s85 )K + r58 S −(1 + s42 )K − r24 S
11 −17 s11 r11
5 3 2 k s a12 K + b12 S −s4 K − (1 + r4 )S −(1 + s11
11 11 11
9 )K − r9 S
6 −9 s6 r6
6 3 2 k s a7 K + b7 S (1 + s7 )K + r7 S −(1 + s4 )K − r46 S
10 10 6

−1 s1 r1
7 3.2 k s a 2 K + b2 S (1 + s52 )K + r25 S s10 10
2 K + (1 + r2 )S
8 −12 s8 r8
8 3 2 k s a9 K + b9 S −s8 K − (1 + r8 )S −(1 + s6 )K − r68 S
8

3 −4 s3 r3
9 3 2 k s a 4 K + b4 S (1 + s74 )K + r47 S −(1 + s3 )K − r3 S
10 −15 s10 r10
10 3 2 k s a11 K + b11 S −s3 K − (1 + r3 )S −(1 + s10
10 10 10
8 )K − r8 S
5 −7 s5 r5 9 9
11 3 2 k s a 6 K + b6 S (1 + s6 )K + r6 S −(1 + s3 )K − r35 S
5

The coefficients verify the relations (in what follows, all sums are
considered modulo 12)
12
 12

ai = bi = −1
i=1 i=1

The partial sums are defined by


b 
n
sba = ai sn = sn1 = ai
i=a i=1

and

b 
n
rab = bi rn = r1n = bi
i=a i=1

To characterize a temperament by a number, James Murray Barbour

(1951) proposed to consider the mean deviation:

a   
12
12 j+6 
md = 1 + 7 sj 
12
j=1
218 A Compendium of Musical Mathematics

where a is defined by
 
37
a = 1200 log 2
213/12
about 14 cents.
Over the years, there have been several solution in pre-20th cen-
tury Europe to temper the fifths. Perhaps, the most successful tuning
was the meantone tuning which appeared sometime around the late
15th century. Pietro Aaron (ca 1485–1545) proposed to diminish the
fifths by a fourth of a syntonic comma except one.
In equal temperament, all fifths are diminished by the twelfth of
a Pythagorean comma.

9.4 Harmonic Metrics

Many authors have proposed new definitions of harmonic distance.

It seems that the mathematical distance on rational numbers are not
well adapted to musical theory. Most of the definitions presented here
have been established by Yves Hellegouarch (1982).
Definition 9.1. The harmonic metric of a rational frequency ratio
p/q (p, q are relatively prime) in the basis b = 2 is
   
p p
d = log2 h
q q

where h(p/q) = max(p, q). For a k-chord of rational frequencies

x = (1, f1 , . . . , fk−1 ), the harmonic metric is defined by

d(x) = log2 max(λ, λf1 , . . . , λfk−1 )

where fi = pi /qi and λ = gcd(qi ).

Remark that the map d is a metric on Q+ \{0}. When the fre-
quencies are not rational, the harmonic metric is extended using the
property

h(xn ) = [h(x)]n ∀n∈N

 Tuning Systems 219

Example 9.1. In the Pythagorean tuning, the major chord x =

(C1 , E, G, C2 ) has the harmonic distance equal to
 
34 3
d(x) = log2 max 1, 6 , , 2
2 2
= log2 max(26 , 34 , 25 .3, 27 )
= log2 (27 ) = 7.
In the 12-tone temperament, for the same chord x = (1, 21/3 ,
27/12 , 2), we have x12 = (1, 24 , 27 , 212 ) and
h(x12 ) = 212 = (h(x))12
So h(x) = 2 and
d(x) = log2 (2) = 1
Definition 9.2. The harmonic gap of a rational frequency p/q (p, q
relatively prime and p, q ≥ 1) is the logarithm sum of its numerator
and denominator:
H (p/q) = log2 p + log2 q
The harmonic distance of two ratios r1 = p1 /q1 and r2 = p2 /q2 is
defined by
δ (r1 , r2 ) = |H (r2 ) − H (r1 )|
Proposition 9.1. The harmonic gap verifies the following
properties:
(1) H (p/q) ≥ 0 for all p/q ∈ [1, 2].
(2) H ((p/q)α ) = αH (p/q) for all p/q ∈ [1, 2], α ∈ R.
(3) H (r1 r2 ) = H (r1 ) + H (r2 ) r1 = p1 /q1 and r2 = p2 /q2 .
Definition 9.3. The harmonic complexity of a frequency ratio
p/q whose decomposition in prime factors has the form p/q =
2α0 pα1 1 · · · pαnn where p1 , . . . , pn are prime numbers greater than 2 is
the sum of the prime factors different of 2:
C(p/q) = p1 + · · · + pn
220 A Compendium of Musical Mathematics

9.5 Continued Fractions

Continued fractions appear, apparently for the first time, in the work
of Hindu mathematician Aryabhata in the late fifth century. A simple
continued fraction is a representation of a real number α given by
1
α = a0 +
1
a1 +
1
a2 +
a3 + · · ·
or in a compact abbreviated notation
α = [a0 ; a1 , a2 , . . .]
with a semicolon to separate the integer part. If the integer part is
null (a0 = 0), the notation α = [a1 , a2 , . . .] is also used. For typo-
graphic convenience, the continued fraction is also written as
1 1 1
a0 + ···
a1 + a2 + a3 +
It is well known that every real number α has a unique continued
fraction expansion. This expansion is finite if α is rational (α ∈ Q)
and infinite if α is irrational (α ∈ R\Q). Let α be the floor function
of α, the greatest integer less or equal α. Define the sequence xn by
the relations x0 = α and
1
xn+1 =
xn − xn 
The numbers an are then the integer part of xn
an = xn 
Define inductively the numbers pn and qn by the relations
p0 = a0, p1 = a0 a1 + 1, pn+1 = pn an+1 + pn−1 , n≥1
q0 = 1, q 1 = a1 , qn+1 = qn an+1 + qn−1 , n≥1
Definition 9.4. The convergents are the rational numbers cn
defined by
pn 1
cn = = [a0 ; a1 , . . . , an ] = a0 +
qn 1
a1 +
..
. 1
+
an
 Tuning Systems 221

Definition 9.5. Given two consecutive convergents pn−1 /qn−1 ,

pn /qn , any fractions of the form
pn−1 + mpn
sn =
qn−1 + mqn
where m is an integer such that 0 ≤ m ≤ an+1 are called the semi-
convergents to α = [a0 ; a1 , a2 , . . .].
Example 9.2. The continued fraction expansion was known by
Euler:
π 1
=
4 1
1+
9
2+
25
2+
2 + ···
(2n + 1)2
··· + 2 +
2 + ···
The convergents to
π = [3; 7, 15, 1, 292, 1, 1, . . .]
are c0 = 3, c1 = 22/7, c2 = 333/106, c3 = 355/113, etc. And the semi-
convergents are 3/1, 13/4, 16/5, 19/6, 22/7, 179/57, 201/64, 223/71,
245/78, 267/85, 289/92, 311/99, 333/106, etc. Note that convergents
are semiconvergents.
Example 9.3. The golden ratio
√
1+ 5
φ= = [1; 1, 1, 1, . . .]
2
has the following convergents: c0 = 1, c1 = 2, c2 = 3/2, c3 = 5/3,
c4 = 8/5, c5 = 13/8, c6 = 21/13, etc.
Proposition 9.2. The continued fractions of a real number α veri-
fies the following properties:
(1) (qn ) is a strictly increasing sequence and tends to +∞.
(2) ∀n ∈ N, pn+1 qn − pn qn+1 = (−1)n .
(3) (c2k ) is an increasing sequence, and (c2k+1 ) is a decreasing
sequence.
(4) ∀k ∈ N, c2k < α < c2k+1 .
222 A Compendium of Musical Mathematics

Proof.
(1) It is easy to see by induction that (qn ) is a strictly increasing
sequence, since q0 = 1 < q1 = a1 , and if qn−1 < qn , we have
qn+1 qn−1
= an+1 + > an+1 > 1
qn qn
(2) Eliminating an between the equations
pn+1 = pn an+1 + pn−1
qn+1 = qn an+1 + qn−1
leads to
pn+1 qn − pn qn+1 = −(pn qn−1 − pn−1 qn−1 )
and since
p1 q 0 − p0 q 1 = 1
we get
pn+1 qn − pn qn+1 = (−1)n
(3) Dividing the previous equation by qn qn+1 , we get
pn+1 pn (−1)n+1
− =
qn+1 qn qn qn+1
or
(−1)n+1
cn+1 − cn =
qn qn+1
Thus, the distance between cn and cn−2 is
pn pn−2
cn − cn−2 = −
qn qn−2
pn−1 pn−2 (−1)n−1
= − +
qn−1 qn−2 qn−1 qn
(−1)n−2 (−1)n−1
= +
qn−2 qn−1 qn−1 qn
 
(−1)n−1 1 1
= −
qn−1 qn qn−2
(−1)n an
=
qn−2 qn−1
 Tuning Systems 223

The even convergents c2k form an increasing sequence and the

odd convergents c2k+1 form a decreasing sequence.


9.6 Best Approximations

Given a real number α, there are two ways to define a best approx-
imation of α. These two kinds of approximation are associated with
two distinct musical measurements and lead to two methods of scale
construction. All results of this section are demonstrated in Khinchin
and Ya (1964).

Definition 9.6. Let α be a real number. The rational number p/q

is the best approximation of the first kind to α if
   
   
α − p  < α − p  (9.1)
 q  q 

for every rational number p /q  = p/q such that 0 < q  ≤ q.

Definition 9.7. Let α be a real number. The rational number p/q

is the best approximation of the second kind to α if
 
|qα − p| < q  α − p  (9.2)

for every rational number p /q  = p/q such that 0 < q  ≤ q.

The best approximation of the second kind is also the best approx-
imation of the first kind, but the converse is false.

Example 9.4. To find q just fifths (3/2) lying in p octaves, one has
to solve the Diophantine equation with p, q = 0:
 q
p 3
2 =
2

This equation has no positive integer solutions. Taking

 3  logarithms,
we have to find the best approximation of α = log2 2 :
  
p 
 − log2 3  = 0
q 2 
224 A Compendium of Musical Mathematics

The continued fraction of α is

 
3
α = log2 = [1, 1, 2, 2, 3, 1, 5, 2, 23, 2, 2, 1, . . .]
2

and the convergents are

1 3 7
[1] = 1, [1, 1] = , [1, 1, 2] = , [1, 1, 2, 2] = ,
2 5 12
24 31 179
[1, 1, 2, 2, 3] = , [1, 1, 2, 2, 3, 1] = , [1, 1, 2, 2, 3, 1, 5] =
41 53 306
which means that we get approximately q just fifths lying in
p octaves: 7 fifths in 12 octaves, 24 fifths in 41 octaves, etc. The
convergents are located as two adjacent sequences (one under α and
the other above):
1 7 31 179 24 3
< < < ··· < α < ··· < < <
2 12 53 306 41 5
The semiconvergents are
1 2 3 5 7 10 17 24 31 55 86 117 148 179
, , , , , , , , , , , , , ,....
2 3 5 7 12 17 29 41 53 94 147 200 253 306
All best approximations of the first kind are semiconvergents, but
not all semiconvergents are best approximations (of the first kind).
For instance, the semiconvergent 10/17 is not the best approximation
to α since
   
   
α − 10  > α − 7 
 17   12 

Theorem 9.1. Among the fractions p/q with q ≤ qn , the closest to

the real number α is the convergent pn /qn .
Proof. Suppose that there exists a fraction p/q such that
   
 p   p 
α −  < α −  n
 q   qn 

and suppose, in a first time, that the fraction p/q is between

cn = pn /qn and cn+1 = pn+1 /qn+1 . The sequence (. . . pn /qn , p/q,
 Tuning Systems 225

pn+1 /qn+1 , . . .) is increasing or decreasing. For a nonnegative inte-

ger N ,
   
 pn p   
 −  = N <  pn+1 − pn  = 1
 qn q  qn q  qn+1 qn  qn qn+1
that is, q > N qn+1 ≥ qn+1 . If the sequence is increasing, we get
pqn > qpn ≥ qn+1 pn+1 ≥ pn+1 qn
thus q > pn+1 , and if the sequence is decreasing, we get
pqn+1 > qpn+1 > qn+1 pn+1
Consequently, p > pn+1 . Now if p/q does not belong to the interval
[cn , cn+1 ] since
1 1
< < ···
qn qn+1 2qn−1 qn
the length of the interval [cn , cn+1 ] is at most the half of the length
of [cn−1 , cn ]. Thus, p/q belongs to [cn−1 , cn ]. From the first part of
the proof, we see that
q > qn and p > pn
So, for a fraction p/q such that
   
   
 α − p  <  α − pn 
 q  qn 
we have q > qn and p > pn . The approximation of the real α by
a convergent is better than the approximation by other fractions
simpler than the convergent. 

Theorem 9.2. If the rational p/q is the best approximation of the

first kind to α, then p/q is a convergent.
Theorem 9.3. The rational p/q is the best approximation of the
second kind to α if and only if p/q is a convergent to α.
Theorem 9.4. For n ≥ 0, we have
 
1  pn  1

< α− <
qn (qn+1 + qn )  qn qn qn+1
226 A Compendium of Musical Mathematics

Proof. Substituting in the expression [a0 , a1 , . . . , an+1 ], an+1 by

the rest αn+1 , the real α is
pn αn+1 + pn−1
α=
qn αn+1 + qn−1
Since αn+1 , pn , qn are positive,
   
   
α − pn  <  pn+1 − pn  = 1
 qn   qn+1 qn  qn qn+1
On the other hand, the distance
   
 pn+1 + pn pn   (pn+1 + pn )qn − pn (qn+1 + qn ) 
   
 qn+1 + qn − qn  =  qn (qn+1 + qn ) 
 
 (−1)n 
=  
qn (qn+1 + qn ) 
1
=
|qn (qn+1 + qn )|
The induction relation induces
qn−1 > qn−2 and qn > 2qn−2
thus
1 1 1
< < < ···
qn qn+1 2qn−1 qn 4qn−1 qn−2
The number α is the sum of the alternate serie
∞  
pn p0  pn pn−1
α = lim = + −
n→∞ qn q0 qn qn−1
n=1

p0 1 (−1)n
= + + ··· + + ···
q0 q0 q1 qn qn+1 

Corollary 9.1. Let α be a real number of convergents an = pn /qn

and m an integer. For n ≥ 1, the following holds:
(1) If 0 ≤ m < 12 an+1 , then
   
   
α − pn  < α − mpn + pn−1 
 qn   mqn + qn−1 
 Tuning Systems 227

(2) If 12 an+1 < m ≤ an+1 , then

   
   
α − mpn + pn−1  < α − pn 
 mqn + qn−1   qn 
Proof.
(1) If 0 ≤ m < 12 an , then (an+1 − 2m)qn ≥ qn ≥ qn−1 . It follows that
qn+1 − qn−1 = an+1 qn ≥ 2mqn + qn−1 . Thus, by the previous
theorem,
   
  1  pn mpn + pn−1 
α − pn  < 1
≤
1
=  −
 qn  qn qn+1 2qn (mqn + qn−1 ) 2 qn mqn + qn−1 
The required inequality follows from the fact that α lies between
pn mpn +pn−1
qn and mqn +qn−1 .

(2) From the inequalities

qn+1 − qn−1 + qn = (an+1 + 1)qn ≤ 2mqn < 2mqn + qn−1

we get

qn+1 + qn < 2(mqn + qn−1 )

Thus, from the previous theorem,

 
1  pn mpn + pn−1  1
 −  =
2 qn mqn + qn−1 2qn (mqn + qn−1 )
 
1  p 
< α − 
n
<
qn (qn + qn−1 ) qn
pn mpn +pn−1
But m ≤ an+1 implies α lies between qn and mqn +qn−1 .
The required inequality follows.


Remark 9.1. As shown by Jack Douthett et al. (1992), the  case

m = 12 an+1 cannot be treated. For the major third, α = log2 54 , they
showed that the semiconvergents 1/4, 2/7, 3/10 and 4/13 between the
convergents 1/3 and 9/28 are worse approximations of the first kind
to α than 1/3, while the semiconvergents 5/16, 6/19, 7/22 and 8/25
between the convergents 1/3 and 9/28 are better approximations.
These results illustrate the two cases of the corollary.
228 A Compendium of Musical Mathematics

Definition 9.8. Let x ∈ [0, 1]\Q be an irrational number and pn /qn

the convergents of x. Assuming x = [0; a1, a2, . . .], the Brjuno func-
tion is
+∞
  
pn−1 − xqn−1
B(x) = |pn−1 − qn−1 | log
n=0
q n x − pn

with by convention (p−1 , q−1 ) = (1, 0).

The first term in the Brjuno function B(x) is log(1/x). The
Brjuno function is extended by periodicity on R\Q.
Proposition 9.3. The Brjuno function satisfies
(1) ∀x ∈ R\Q, B(x + 1) = B(x),
(2) ∀x ∈ [0, 1]\Q, B(x) = log(1/x) + xB(1/x).
Definition 9.9. The Diophantine order of x is the number

τ (x) = lim sup τn (x)

n→∞

where numbers τn (x) are defined by

 
 
x −  = 1
p n
 qn  qnτn (x)
 
 pn 
1
Remark 9.2. If x is an irrational number, then x − qn , thus
<
qn2
τn (x) > 2. The Diophantine order of an irrational number is then
greater or equal to 2.

9.7 Musical Scale Construction

Given a rational interval, such as a fifth (3/2), one way to construct

a musical scale is to repeat this interval until it reaches in the neigh-
borhood of the octave, as it is done in the Pythagorean tuning. Then
to find q just fifths (3/2) lying in p octaves, one has to solve the
Diophantine equation with p, q = 0:
 q
p 3
2 =
2
 Tuning Systems 229

As we have seen, this equation has no positive integer solutions. This

procedure can be generalized to any rational number a/b, where a
and b are coprimes.
Theorem 9.5. The sequence of the number N of notes in the best
approximation by an N-EDO of a tuning system constructed by the
repetition q times of a rational interval a/b in p octaves is given by
the denominators of the convergents of the irrational number:
a
α = log2
b
Example 9.5. For the fifth 3/2, the denominators of the convergents
are the numbers 12, 41, 53, 306, etc. The first approximations were
known by Pythagoras (12), Jankó (41) and Hölder (53). For the
major third 5/4, the sequence is 3, 28, 59, 146, etc. And for the
minor third 6/5, we get 3, 4, 19, 422, 1707, etc.
Example 9.6. In the classical meantone temperament, designed
by Pietro Aaron (ca 1485-1545), the fifths (3/2) are tempered by
1/4 of a syntonic comma (81/80). Thus, the ratio of the meantone
fifth is
 
3 81 −1/4
Q= = 51/4
2 80
and its value is 697 cents. The Diophantine equation

2q = Qq = 5q/4

has no trivial solution. It leads to the irrational number α = 14 log2 (5)

whose convergents are 3/5, 4/7, 7/12, 11/19, 18/31, 101/174, etc.
It follows that the best approximation by an N-EDO is obtained for
N = 12, 19, 31, etc. Some other approximations are based on the
approximation of the fifth Q. For instance, Verheyen (as given by
Simon Stevin in 1884) used 51/4 ≈ 195/128 and Gibelius (1666) used
51/4 ≈ 160/107.
Conversely, given an N -EDO, one can build the best approxima-
tion tuning using rational numbers. For each frequency αk = 2k/N ,
with k = 0, 1, . . . , N − 1, the computation of the convergents ck,n
of αk leads to a sequence of rationals of 1200 log 2 (ck,n ) cents. For
230 A Compendium of Musical Mathematics

instance, we keep the first convergent ck such that its approximated

ratio is less than 0.5 cents:
 
 k 
1200 log2 (ck ) −  < 1/2
N

Example 9.7. The computation of the approximate 1200-EDO

leads to c0 = 1, c1 = 1731/1730, c2 = 866/865, c3 = 577/576, etc.

Example 9.8. Superpartial ratios (i.e. ratios of the form (n + 1)/n)

have an important property: superpartial ratios are Hellegouarch
commas. They are ratios of the first just intervals (2/1, 3/2, 4/3, 5/4,
6/5, etc.) and also they are the best approximations of microtones.
In the sequence of convergents of the approximation of a microtone
of frequency 21/N in the N -EDO, there is always a superpartial ratio.
The following table shows the superpartial ratios in the computation
by continued fractions of the microtones (second column), the num-
ber of degrees N (N -EDO, first column) and the values in cents of
the microtones (third column):

N Ratios Cents 19 28/27 63 54 79/78 22

10 15/14 120 24 35/34 50 60 87/86 20
12 17/16 105 31 45/44 39 66 95/94 18
17 25/24 71 36 52/51 33 72 104/103 17
18 26/25 68 53 77/76 23 96 138/137 13

9.7.1 Euler–Fokker genera

In the 18th century, Swiss mathematician Leonhard Euler (1707–
1783) had made up a new procedure to carry out tuning systems
called musical genus (genus musicum) from a simple mathematical
formula (Tentamen Novae Theoriae Musicae, 1739). Starting from
an original pitch level n, Euler appoints intervals by their rank in
the partials decomposition of a fundamental pitch: 3 for the fifth,
5 for the third, 7 for the seventh, etc. The Euler genera given by the
formula

3a · 5b · 7c · · · n

describe all products of the form 3a × 5b × · · · × n.

 Tuning Systems 231

Example 9.9. For instance, Euler genera of type 32 · 53 · n are the

set of all products of the form 3a · 5b · n, where a = 0, 1, 2 and
b = 0, 1, 2, 3. So, one gets the set of numbers n, 3n, 5n, 9n, 15n, 25n,
45n, 75n, 125n, 225n, 375n, 1125n. If one gives the value of n, for
example, n = 1/5, the sequence becomes 1/5, 3/5, 1, 9/5, 3, 5, 9,
15, 25, 45, 75, 225. Normalizing in the octave [1, 2] by introducing
powers of 2, one gets a complete tuning system: C (1), D (9/8, 204
cents), D+ (75/64, 275 cents), D (6/5, 316 cents), E (5/4, 386 cents),
F (45/32, 590 cents), G (3/2, 702 cents), G+ (25/16, 773 cents),
G (8/5, 814 cents), A+ (225/128, 977 cents), A (9/5, 1018 cents)
and B (15/8, 1088 cents).

Example 9.10. Euler genera could describe many well-known tem-

peraments. The genus 36 · n is the Eratosthenes diatonic system
described by Ptolemaus. The genus 311 · n is the Pythagorean tem-
perament built on the fifths ascending (if n = 311 ) and descending
(if n = 1/311 ). The genus 33 · 52 · n corresponds to the Zarlinian
temperament.

In the 20th century, this procedure has been widely studied

by Adriaan Fokker (1889–1972) in relation with lattices properties.
Based on Christiaan Huygens works (see, for instance, Caddeo et al.,
2021), Fokker reintroduced the 31-EDO. From the Euler expression
3p ·5q ·7s , he draws in a multidimensional space, p fifths in the x direc-
tion, q thirds in the y direction over the fifths and s sevenths in the
z direction.

9.7.2 Farey tunings

Farey tunings could be defined from Farey sequences. Introduced by
Haros in 1802 and independently by John Farey in 1816, the Farey
sequence is defined as follows.

Definition 9.10. The Farey sequence Fn of order n is the increasing

sequence of irreductible fractions p/q (i.e. gcd(p, q) = 1) between 0
and 1 whose denominators do not exceed n. The Farey tuning T Fn
is the shifted Farey sequence, T Fn = Fn + 1.
232 A Compendium of Musical Mathematics

Farey sequences are built recursively from the set

F1 = {0, 1}
by inserting mediants of each neighboring pairs of rational numbers.
If p/q and p /q  are such pairs, the mediant is defined by
p + p
q + q
Example 9.11. The first Farey sequences are
0 1 1 0 1 1 2 1
F2 = , , ⊂ F3 = , , , , ⊂ F4
1 2 1 1 3 2 3 1
0 1 1 1 2 3 1
= , , , , , ,
1 4 3 2 3 4 1
The main properties of the Farey sequences are as follows:
Proposition 9.4.
(1) If p1 /q1 , p2 /q2 are two consecutive ratios p1 /q1 < p2 /q2 , then
|p1 q2 − q1 p2 | = 1
(2) If p1 /q1 , p2 /q2 , p3 /q3 are three consecutive ratios p1 /q1 <
p2 /q2 < p3 /q3 , then
p2 p1 + p3
=
q2 q1 + q3
(3) The number of terms in the Farey sequence of order n (i.e. the
cardinal of the set Fn ) is

n
|Fn | = 1 + ϕ(k) (9.3)
k=1

where ϕ(k) is the Euler totient function.

Example 9.12. Ivan Wyschnegradsky (1996) in his research on
rhythmical figures of ultrachromatism has considered the 115 frac-
tions whose numerators and denominators are less or equal 13. Using
the formula 9.3, we see that for n = 13, the Farey set F13 has car-
dinal 59. Thus, the set A = F13 \{0, 1} has cardinal 57. The set
 Tuning Systems 233

considered by Wyschnegradsky is the set A ∪ A ∪ {1}, with A the

set of the reverse fraction of A, A = {p/q, q/p ∈ A}. It follows
   
A ∪ A ∪ {1} = 1 + |A| + A  = 1 + 2 |A| = 1 + 2 × 57 = 115

since A and A are distinct.

The Farey sequences (and then the Farey tunings) are related
to Ford circles. The Ford circle C(p, q) is the circle centered at
(p/q, ±1/(2p2 )) and of radius 1/(2q 2 ). Ford circles provide an inter-
esting sphere packing. If we put the Farey ratios pi /qi on the real line,
we could draw the circles C(pi , qi ). If p1 /q1 , p2 /q2 , p3 /q3 are three
consecutive terms of the Farey sequence, then the circles C(p1 , q1 )
and C(p2 , q2 ) are tangent at the point a1 given by
 
p2 q1 1
a1 = + ,
q2 q2 (q22 + q12 ) q22 + q12
and the circles C(p2 , q2 ) and C(p3 , q3 ) are tangent at the point a2
given by
 
p2 q3 1
a2 = + ,
q2 q2 (q22 + q32 ) q22 + q32
The point a1 lies on the circumference of the circle built on the
two points of the real line (p1 /q1 , 0) and (p2 /q2 , 0). The point a2
belongs to the circle built on (p2 /q2 , 0) and (p3 /q3 , 0). If we plot
all the fractions on the x-axis and their corresponding Ford circles
generated from a Farey series of order n, they form a set of tangent
circles.

9.7.3 Partch odd-limit

Definition 9.11. Let n be a positive odd number. The n-limit sys-
tem contains all rational number such that the largest odd num-
ber that divides either the denominator or numerator is not greater
that n.
If n is a prime number p, the p-prime-limit contains all rational
numbers that can be factored using primes no greater than n.
Proposition 9.5. The p-prime-limit tuning is a subgroup of the
additive group Q+ .
234 A Compendium of Musical Mathematics

Example 9.13. The Partch 7-limit temperament is built on num-

bers 3, 5, 7 and leads to a system of 13 degrees:

{1, 8/7, 7/6, 6/5, 5/4, 4/3, 7/5, 10/7, 3/2, 8/5, 5/3, 12/7, 7/4}

Example 9.14. The Partch 9-limit is built on the numbers 3, 5, 7, 9

and leads to the 19 degrees temperament:

{1, 10/9, 9/8, 8/7, 7/6, 6/5, 5/4, 9/7, 4/3, 7/5, 10/7, 3/2,
14/9, 8/5, 5/3, 12/7, 7/4, 16/9, 9/5}

Example 9.15. The Partch 11-limit is the temperament of 29

degrees:

{1, 12/11, 11/10, 10/9, 9/8, 8/7, 7/6, 6/5, 11/9, 5/4, 14/11, 9/7,
4/3, 11/8, 7/5, 10/7, 16/11, 3/2, 14/9, 11/7, 8/5, 18/11, 5/3,
12/7, 7/4, 16/9, 9/5, 20/11, 11/16}

9.8 Three-Gap Theorem and Cyclic Tunings

Formerly known as the Steinhaus conjecture, the three-gap theorem

has been proved by Vera T. Sós (1957, 1958), János Surányi (1958)
and many others. A simple proof has been given by Frank Liang
(1979).
Theorem 9.6 (Three-gap theorem). Let N points be placed on a
circle, at angles of α, 2α, 3α. . . from a given starting point, then
for all irrational α and natural N, the points partition the circle into
gaps of at most three different lengths. In other words, there will be
at most three distinct distances between pairs of points in adjacent
positions around the circle. Moreover, when there are three distances,
the largest of the three always equals the sum of the other two. Unless
α is a rational multiple of π, there will also be at least two distinct
distances.
In music theory, the three-gap theorem has the following conse-
quence: if a tuning system in which one considers interval modulo the
octave is generated by a given musical interval w, then there are at
most three different intervals between consecutive tones of the scale.
 Tuning Systems 235

9.8.1 Cyclic tunings

Many irrational numbers have been used in artistic forms for their
harmonic proportion, beauty and strange fascination. Cyclic tunings
(called Moments of symmetry by Ervin Wilson, see Narushima, 2016)
are made by the repetition of any number w, with the following pro-
cedure. Write the sequence of powers of the number w, for instance,
{w−k , w−k+1 , . . . , w−1 , 1, w, . . . , wk }
and normalize each ratio by a power of 2 to put them in the same
octave [1, 2]. This leads to a new tuning system which have many
common properties. The most famous examples of the cyclic tunings
are the Pythagorean tunings based on the pure fifth (w = 3/2) and
the golden scale based on the golden fifth (a fifth which depends
on the golden ratio, solution of the equation φ2 = φ + 1). One can
also consider tunings generated by any irrational number such as π =
3.14159 . . . , e = 2.71828 . . ., the Euler constant γ = 0.5772156649 . . .,
the Catalan number C = 0.9159655942 or the Feigenbaum number
F = 4.6692106 . . .. As an example, we build the Pythagorean tunings.
Tuning systems based on a number w are obtained in the same way.
Pythagorean Tunings: The simplest Pythagorean scale (k = 1)
based on the repetition of the pure fifth (c0 = 3/2 = 702 cents)
is the system {4/3, 1, 3/2} composed of three notes {F, C, G} and
reorganized {C, F, G} is described by the word
P3 = c1 c2 c1
where c1 = 4/3 = 498 cents and c2 = c0 /c1 = 9/8 = 204 cents. The
next scale (with k = 2) is made from the fifths {B , F, C, G, D } and
composed of two intervals c2 and c3 = c1 /c2 = 32/27 = 294 cents.
The scale is made of five notes {C, D, F, G, B } and represented by
the word
P5 = c2 c3 c2 c3 c2 = (c2 c3 )2 c2
For k = 3, we get the seven degrees Pythagorean scale {C, D, E ,
F, G, A, B } represented by the word
P7 = c2 c4 (c2 )3 c4 c2
in which the two constituent intervals are related to the previous
one by the relation c4 = c3 /c2 = 256/243. This system has only two
236 A Compendium of Musical Mathematics

constituent intervals c2 and c4 . If we continue the algorithm with

k = 4, we construct the scale {C 1, D 9/8, E 32/27, E 81/84,
F 4/3, G 3/2, A 128/81, A 27/16, B 16/9 } which is described by
the word
P9 = a(bcb)a(bcb)a
where the intervals a, b and c are given by a = 9/8, b = 256/243
and c = 2187/2048. We see that this scale has three constituent
intervals instead of two. The first scale which contains all the previous
scales and has only two intervals is the scale P17 , involving the two
intervals 256/243 and b = 312 /219 . The following table shows the
next steps. The first column gives the number of degrees in the scale,
the second column is the structure of the tuning and the third and
fourth columns give the ratio of a and b. The fifth column gives the
ratios of the constituent intervals in cents:

3 aba 4/3 9/8 498, 204

5 ababa 9/8 32/27 204, 294
7 aba3 ba 9/8 256/243 204, 90
17 12a, 5b 256/243 312 /219 90, 23
29 17a, 12b 312 /219 227 /317 23, 67
41 29a, 12b 312 /219 246 /329 23, 43
53 41a, 12b 312 /219 265 /341 23, 20

9.8.2 Ervin Wilson’s CPS

Ervin Wilson has proposed a method for building temperaments
named Combination Product Sets (CPSs). This procedure has been
commented by many authors; see Narushima (2016). Starting from a
set of generators G = [g1 , g2 , . . . , gn ], the method consists in forming
all
 the possible factors by pointwise multiplication of p elements. The
n
p factors are ordered and doubles are eliminated. This leads to a
set A. The elements (x, y) of the Cartesian product A × A are then
divided by themselves x/y and adjusted by a power of 2 to remain
in the octave [1, 2]. The reordered set or CPS is denoted by
p n [g1 , g2 , . . . , gn ]
 Tuning Systems 237

Example 9.16. For instance, the hexany

2 4 [1, 3, 5, 7]
is built on the six products 1 · 1, 1 · 3, 1 · 5, 1 · 7, 3 · 5, 5 · 7 normalized
by a power of 2, as shown in the following table:
1/1 7/6 5/4 35/24 5/3 7/4
1/1 21/20 6/5 7/5 3/2 7/4
1/1 15/14 5/4 10/7 3/2 12/7
1/1 7/6 4/3 7/5 8/5 28/15
1/1 8/7 4/3 10/7 5/3 40/21
1/1 8/7 6/5 48/35 8/5 12/7
Eliminating the doubles, we get a 19-note temperament.
Example 9.17. Many CPSs have a great number of degrees. But
the CPS 3 9 [1, 2, . . . , 9] leads to a system of 15 notes, and the
CPS 4 8 [1, 2, . . . , 8] gives a 12-note temperament.

9.9 Tuning Theory

Most of the results of tuning theory have been established by Yves

Hellegouarch (1982), Kassel and Kassel (2010), Carl Lumma (2001)
and many others microtonalists. After reviewing Hellegouarch’s
approach to commas, we present a unified theory of tuning systems.

9.9.1 Hellegouarch’s commas

All the results of this section have been established by Yves
Hellegouarch (1982). Let p1 , p2 , pn be n prime numbers with
p1 < p2 < · · · < pn . We denote G = p1 , . . . , pn  the subgroup of
Q∗+ generated by the prime numbers pj .
Definition 9.12. A Hellegouarch comma of the group G (or a best
approximation of 1 in G) is a number a = 1 verifying
b ∈ G\{1} and |log b| < |log a| ⇒ h(b) > h(a)
where h(p/q) = max(p, q) is the height of the rational number p/q.
238 A Compendium of Musical Mathematics

Remark that if a is a comma of a subgroup H of G, a is

not necessarily a comma of G. For example, 28 /35 is a comma of
H = 2, 3 but not a comma of G = 2, 3, 5. If a = px1 1 · · · pxnn is a
comma, then gcd(x1 , . . . , xn ) = 1.
Proposition 9.6. Let a be a positive rational number, then (a+1)/a
is a comma of G.
Proof. Let p/q > 1 be a positive rational. One must show that if
p/q > 1 and log(p/q) < log((a + 1)/a), then p > a + 1. But since
the function x → (x + 1)/x is decreasing on ]0, +∞[ and (q + 1)/q ≤
p/q ≤ (a + 1)/a, we have q > a. The inequalities p ≥ q + 1 ≥ a + 1
lead to the result. 
Theorem 9.7. Let p, q be two prime numbers with p < q. The
commas of the group G = p, q are the convergents of the irrational
number α = log p/ log q.
Proof. We must show that all commas are convergents. Let a =
px q y be a comma. Then x and y have opposite signs and |x/y| is the
best approximation of α. Thus, a is a convergent. Conversely, let’s
define the sequence (an ) by
q rn−2
r−1 = p, r0 = a0 , rn =
p (rn−1 )an
Since p < q, the number a0 exists and r0 is the first convergent of α.
There exits some numbers an ≥ 1 such that
an an +1
rn−1 < rn−2 < rn−1
because rn−1 < rn−2 and (log(rn−1 ), log(rn−2 )) are linearly indepen-
dent on Q. The number rn is a comma. Let b = px q y such that
   
logp (b) < logp (rn )
that is,
   
x + y logp (q) < xn + yn logp (q)
and
(xn , yn ) = ((−1)n−1 pn , (−1)n qn )
Since pn /qn is the best approximation of log p (q), we have |y| < qn
and |x| > pn . Consequently, x and y have opposite signs and then
h(b) > h(rn ). 
 Tuning Systems 239

Definition 9.13. Let G be the group G = p, q and rn a comma

of G. The group G/ rn  is isomorphic to Z and defines natural scales.
For p = 2 and q = 3, the scales are called Pythagorean scales.

Example 9.18. For the first convergents, we recover the classical

tunings Hn = 2, 3 / rn  . For r1 = 4/3, H1 = {1, 2}. For r2 = 9/8,
H2 = {1, 3/2, 2}. For r3 = 28 /35 , H3 = {1, 9/8, 4/3, 3/2, 16/9, 2}.
For r4 = 312 /219 , H4 is the 12-Pythagorean temperament. For
r5 = 265 /341 , H5 is the Jankó temperament of 41-degrees. For
r6 = 353 /284 , H6 is the Mercator temperament of 53 degrees.

9.9.2 Tuning systems

Let G be a finitely generated multiplicative group of the positive
rational Q∗+ .

Definition 9.14. A degree of G is a surjective homomorphism

deg : G → Z.

Definition 9.15. The height of an element x ∈ Q∗+ is the positive

integer h(x) = max(p, q), where x = p/q and p, q are coprime.

Definition 9.16. An element r ∈ G is a comma if deg(r) = 0.

Example 9.19. Let G be the group G = 2, 3, 5 and deg be the

homomorphism:

deg(2x 3y 5z ) = 12x + 19y + 28z

The element r = 81/80 is a comma, since −12 × 4 + 19 × 4 − 28 = 0.

The following definition is Hellegouarch’s definition of natural

scales.

Definition 9.17. Given a natural finitely generated subgroup G of

Q∗+ , an integer N ≥ 2 and a degree deg : G → Z, a tuning system is
a subset Γ ⊂ G of N elements such that the restriction of the degree
to Γ is a bijection onto Γ and each element x ∈ Γ is of minimal height
in the set deg−1 (deg(x)). The integer N is called the rank of G or
the dimension of the tuning system.
240 A Compendium of Musical Mathematics

Since there is no torsion element, G is a free Abelian group of finite

rank. There are finitely many positive real numbers (g1 , g2 , . . . , gN )
such that any element g of G can be written as
xN
g = g1x1 g2x2 · · · gN
for some x1 , . . . xN ∈ Z. The degree deg(g) is then determined by its
value on the basis {g1 , g2 , . . . , gN } of G. The surjectivity of deg(g)
means that deg(g1 ), . . . , deg(gN ) are coprimes and generate Z.
With the notations of the quantum mechanics, the degree could
be written as a bra vector :
t| = deg(g1 ), . . . , deg(gN )|
It is sometimes called a tuning vector. In the same way, the exponents
of the elements of G could be written as a ket vector :
|x = |x1 , . . . , xN 
The tuning vector t = (deg(g1 ), . . . , deg(gN )) ∈ RN maps each ele-
ment x ∈ ZN to

N
t | x = deg(g1 ), . . . , deg(gN )| x1 , . . . , xN  = deg(gk )xk
k=1
which is the pitch height of x.
Example 9.20. When (g1 , g2 , . . . , gN ) are prime numbers gj = pj ,
the degree is often used to be deg(p1 ) = N and
deg(pj ) = round(N log2 (pj )) for j = 2, . . . , N
For instance, for the six first primes {2, 3, 5, 7, 11, 13}, we get for
N = 7, 7, 11, 16, 20, 24, 26| , for N = 12, 12, 19, 28, 34, 42, 44| and
for N = 31, 31, 49, 72, 87, 107, 115| .
Example 9.21. Let G be the free group G = 2, 3, 5 generated by
three generators and choose the degree as follows:
deg(2x 3y 5z ) = 12x + 19y + 28z
Assume that the generators 2, 3, 5 are tempered to 2 → t1 , 3 → t2
and 5 → t3 by the syntonic comma t−4 4 −1 = 1 and the diesis
1 t2 t3
t31 t42 t−4
3 = 1. The two kets |−4, 4, −1 and |3, 4, −4 have a scalar
product of zero:
12, 19, 28| x, y, z = 0
 Tuning Systems 241

A given element 2x 3y 5z of G is tempered to

tx1 ty2 tz3 = α12x19y+28z = αdeg(g)
with t1 = α12 , t2 = α19 , t3 = α28 .
(1) If we choose t1 = 2, then we get the usual 12-EDO, since
α = 21/12 . By temperamentation, we have mapped the octave to the
octave 2 → 2, the perfect fifth to the tempered fifth 3/2 → 27/12 and
the pure major third to the tempered third 5/4 → 24/12 .
(2) If we choose t1 = (3/2)12/7 (an octave of about 1203 cents),
then we get Serge Cordier’s tempered scale (Cordier, 1982) with a
semitone of α = (3/2)1/7 . Octaves are slightly increased, but all fifths
are pure (3/2).
In Kassel and Kassel (2010), Adrien and Christian Kassel have
slightly modified Hellegouarch’s definition of a scale by adding the
constraint that all frequencies are lying in the octave.
Definition 9.18 (Kassel, 2010). Given a natural finitely gener-
ated subgroup G of Q∗+ , an integer N ≥ 2 and a degree deg : G → Z,
such that deg(2) = N , a tuning system (called scale) is a sub-
set Γ ⊂ G of N elements, Γ = {1, x1 , x2 , . . . , xN −1 } such that for
all k = 1, . . . , N − 1, the element xk ∈ G satisfies the following
conditions:
(1) 1 < xk < 2.
(2) deg(xk ) = k and xk is an element of minimal height of the set
Γk = {x ∈ G ∩ (1, 2), deg ∩(x) = k}.
They showed that if G is a free Abelian group of rank ≥ 3, then
scales exist for all choices of the degree. But when the rank is 2,
scales do not always exist.
Theorem 9.8 (Kassel, 2010). If G = 2, 3 is the Pythagorean
group, freely generated by 2 and 3, and if p = deg(2), q = deg(3)
such that p > q ≥ 2, and
 
 
log2 (3) − p  < 1
 q  q2
then there exists a unique scale, which is the Pythagorean scale com-
posed of successive perfect fifths.
The proof of this theorem is given in Kassel and Kassel (2010).
242 A Compendium of Musical Mathematics

9.10 Exercises

Exercise 9.1
Compute the first convergents of ln(6/5)/ ln(2) and show that
 6 19
5 ≈ 25 . Evaluate the difference in cents.
Exercise 9.2
Compute the first convergents of ln(5/4)/ ln(2) and show that
 5 28
4 ≈ 29 . Evaluate the difference in cents.
Exercise 9.3
Compute the tuning induced by the Farey sequence F6 .
Exercise 9.4
Compute the Euler–Fokker tuning 34 · 5.
Exercise 9.5
Study the cyclic tuning for the major third w = 5/4.
Exercise 9.6
Study the cyclic tuning for the minor third w = 6/5.
Exercise 9.7
Let G be the free group G = 2, 3, 5 and define the degree by
deg(2x 3y 5z ) = 7x + 11y + 16z
Compute the Kassel–Hellgouarch tuning. Find two kets |x, y, z solu-
tions of
7, 11, 16 |x, y, z = 0
and compute the associated interval 2x 3y 5z .
Exercise 9.8
Let us consider the degree
deg(2x 3y 7z ) = 7x + 11y + 20z
Compute the Kassel–Hellgouarch tuning. Find two kets |x, y, z
solutions of
7, 11, 20 |x, y, z = 0
and compute the associated interval 2x 3y 7z .
 Tuning Systems 243

References

Barbour, J.M. (1951). Tuning and Temperament. A Historical Survey, Michigan

State College Press, East Lansing. Reprint by Dover Publications, Mineola,
New York.
Caddeo, R., Hascher, X., Jedrzejewski, F. and Papadopoulos, A. (2021). Christi-
aan Huygens, Écrits sur la musique et le son, Hermann, Paris.
Carey, N. (2007). Coherence and sameness in well-formed and pairwise well-
formed scales, Journal of Mathematics and Music 1(2), 79–98.
Cordier, S. (1982). Piano bien tempéré et justesse orchestrale, Buchet/Chastel,
Paris.
Douthett, J., Entringer, R. and Mulhaupt, A. (1992). Musical scale construction:
The continued fraction compromise, Utilitas Mathematica 42, 97–113.
Hellegouarch, Y. (1982). Scales, Mathematical Reports of the Academy of Science
(The Royal Society of Canada) 4(5) [Erratum (1983), 5(2)].
Jedrzejewski, F. (2002). Mathématiques des systèmes acoustiques, Tempéraments
et modèles contemporains, L’Harmattan, Paris.
Jedrzejewski, F. (2014). Dictionnaire des musiques microtonales (1883–2013),
2nd edn., L’Harmattan, Paris.
Jedrzejewski, F. and Papadopoulos, A. (2021). Joseph Sauveur : Écrits sur la
musique et l’acoustique, Hermann, Paris.
Kassel, A. and Kassel, C. (2010). On Hellegouarch’s definition of musical scales,
Journal of Mathematics and Music 4(1), 31–43.
Khinchin, I.Y. (1964). Continued Fractions, The University of Chicago Press,
New York.
Liang, F. (1979). A short proof of the 3d distance theorem, Discrete Mathematics
28(3), 325–326.
Lumma, C. (2001). Tuning theory. http://lumma.org/tuning/gws/home.htm.
Narushima, T. (2016). Microtonality and the Tuning Systems of Erv Wilson:
Mapping the Harmonic Spectrum, Routledge, New York.
Sós, V.T. (1957). On the theory of diophantine approximations, I., Acta Mathe-
matica Academiae Scientiarum Hungarica 8, 461–472.
Sós, V.T. (1958). On the distribution mod 1 of the sequence nα, Annales
Universitatis Scientiarum Budapestinensis de Rolando Eotvos Nominatae.
Sectio Mathematica 1, 127–134.
Surányi, J. (1958). Über dei Anordnung der Vielfachen einer reelen Zahl mod 1,
Annales Universitatis Scientiarum Budapestinensis de Rolando Eotvos
Nominatae. Sectio Mathematica 1, 107–111.
Wyschnegradsky, I. (1996). La loi de la pansonorité, (1953), texte établi et annoté
par Franck Jedrzejewski avec la collaboration de Pascale Criton, postface
de Franck Jedrzejewski, Genève, Contrechamps.
This page intentionally left blank
 Solutions to Exercises

Solution 1.1
For N = 6, the divisors are {1, 2, 3, 6}. For |HA | = 6, we get the
set E1 = {0, 1, 2, 3, 4, 5}. For |HA | = 3, E2 = {0, 2, 4}. And for
|HA | = 2, E3 = {0, 3} and E4 = E3 ∪ T1 (E3 ) = {0, 1, 3, 4}. All others
combinations are transpositions of these sets.
Solution 1.2  
For intervals (k = 2), 62 = 15 = 2 × 6 + 1 × 3. There are 1 interval
of limited transposition (the set {0, 3} of the previous exercise) with
3 transpositions, and 2 others
 intervals with 6 transpositions.
For 3-chords (k = 3), 63 = 20 = 3 × 6 + 1 × 2. There are
four chords up to transpositions, one of which is a set of limited
transposition (the set {0, 1, 3, 4}).
Using the polynomial (1.1), we get

PC6 (x) = 1 + x + 3x2 + 4x3 + 3x4 + x5 + x6

This leads to the following table:

k 1 2 3 4 5 6
k-chords 1 3 4 3 1 1
SLT 0 1 1 1 0 1

245
246 A Compendium of Musical Mathematics

Solution 1.3
For N = 8, the divisors are {1, 2, 4, 8}. For |HA | = 8, we get the
set E1 = {0, 1, 2, 3, 4, 5, 6, 7}. For |HA | = 4, E2 = {0, 2, 4, 6}. And
for |HA | = 2, E3 = {0, 4}, E4 = E3 ∪ T1 (E3 ) = {0, 1, 4, 5}, and
E5 = E3 ∪ T1 (E3 )∪ T2 (E3 ) = {0, 1, 2, 4, 5, 6}. All others combinations
are transpositions of these sets.

Solution 1.4  
For intervals (k = 2), 82 = 28 = 3 × 8 + 1 × 4. There are 1 interval
of limited transposition (the set {0, 4} of the previous exercise) with
4 transpositions, and 3 others
 8  intervals with 8 transpositions.
For 3-chords (k = 3), 3 = 56 = 7 × 8. There are 7 chords up to
transpositions and no set of limited transposition.
For 4-chords (k = 4), 84 = 70 = 8 × 8 + 1 × 2 + 1 × 4. There
are 10 chords up to transpositions, two of which are sets of limited
transposition (the set {0, 2, 4, 6} has 2 transpositions and the set
{0, 1, 4, 5} has 4 transpositions).
Using the polynomial (1.1), we get

PC8 (x) = 1 + x + 4x2 + 7x3 + 10x4 + 7x5 + 4x6 + x7 + x8

This leads to the following table:

k 1 2 3 4 5 6 7 8
k-chords 1 4 7 10 7 4 1 1
SLT 0 1 0 1 0 1 0 1

Solution 1.5

PD6 (x) = 1 + x + 3x2 + 3x3 + 3x4 + x5 + x6

Solution 1.6

PD8 (x) = 1 + x + 4x2 + 5x3 + 8x4 + 5x5 + 4x6 + x7 + x8

Solution 1.7
The inverse of M3 is M3 .
 Solutions to Exercises 247

The M3 permutation is
 
0 1 2 3 4 5 6 7
M3 =
0 3 6 1 4 7 2 5
= (1, 3)(2, 6)(5, 7)
The order of the group is 32.
Solution 2.1
For two points (x, n) and (y, m) of Z12 ×Z2 , we introduce the interval
function
int ((x, n), (y, m)) = (2y + m) − (2x + n)
It is easy to verify that (Z12 × Z2 , Z24 , int) is a GIS.
Solution 2.2
Using the properties of the injection number, for all bijective func-
tion f
inj(X, Y )(f ) = inj(Y, X)(f −1 )
we already get the result since M3 is the inverse of M3 and M5 is the
inverse of M5 . Moreover,
inj(M3 (X), Y )(M3 ) = inj(X, Y )(M3 M3 ) = inj(X, Y )(id) = card(X ∩ Y )
and
inj(X, M5 (Y ))(M3 ) = inj(X, Y )(M5 M5 ) = inj(X, Y )(id) = card(X ∩ Y )

Solution 2.3
The proofs are the same as in the previous exercise. Using the prop-
erties of the injection number, for all bijective function f
inj(X, Y )(f ) = inj(Y, X)(f −1 )
we already get the result since M5 is the inverse of M7 . Moreover,
inj(M5 (X), Y )(M7 ) = inj(X, Y )(M7 M5 ) = inj(X, Y )(id) = card(X ∩ Y )
and
inj(X, M7 (Y ))(M7 ) = inj(X, Y )(M5 M7 ) = inj(X, Y )(id) = card(X ∩ Y )
248 A Compendium of Musical Mathematics

Solution 2.4
By definition,

ibang(X, Y )(n) = 1X (k)1Y (n + 5k)
k

Assuming j = n + 5k, we see that 25k = k = 5(j − n) mod 12, thus

 
ibang(X, Y )(n) = 1X (5(j − n))1Y (j) = 1M5 Tn (X) (j)1Y (j)
j j

= card (Tn M5 (X) ∩ Y )

For X = {0, 1, 4} and Y = {0, 2},
ibang(X, Y ) = (1, 0, 1, 0, 1, 0, 1, 1, 0, 1, 0, 0)
We have
ibang(Ti (X), Y )(n) = card (Tn M5 Ti (X) ∩ Y )
= card (Tn+5i M5 (X) ∩ Y )
= ibang(X, Y )(n + 5i)
For the second equation, we have
ibang(X, Ti (Y ))(n) = card (Tn M5 (X) ∩ Ti (Y ))
 
= card Tn M5 Ti−1 (X) ∩ Y
= card (Tn M5 T−i (X) ∩ Y )
= card (Tn−5i M5 (X) ∩ Y )
= ibang(X, Y )(n − 5i)
And for the last one, we have
ibang(Ti (X), Ti (Y ))(n) = card (Tn M5 Ti (X) ∩ Ti (Y ))
 
= card Tn M5 Ti Ti−1 (X) ∩ Y
= card (Tn M5 (X) ∩ Y )
= ibang(X, Y )(n)
 Solutions to Exercises 249

Solution 2.5
We have
 
FTn (X) (x) = e−2iπkx/N = e−2iπ(j+n)x/N
k∈Tn (X) j∈X

= e−2iπnx/N e−2iπjx/N = e−2iπnx/N FX (x)
j∈X

Solution 2.6
By calculating the table of differences, we get

iv(X) = ifunc(X) = (5, 2, 2, 2, 2, 2, 2, 2, 2, 2) = iv(Y )

Since their interval vectors are equal, the two sets are homometric.
The Patterson polynomial of X is

PX (x) = 5 + 2x + 2x2 + 2x3 + 2x4 + 2x5 + 2x6 + 2x7 + 2x8 + 2x9

= 5 + 2(x + x2 + · · · + x9 )

Using Theorem 2.26, we see that X is a (10, 5, 2)-difference set. The

same result holds for Y , since X and Y have the same Patterson
polynomial.

Solution 2.7
Since X and Y are two complementary sets, they are homometric.
Their interval content is ic(X) = 433221. By multiplying by
5 mod 12, the sets M5 X = {0, 1, 3, 5, 6, 10} and M5 Y = {0, 3, 5,
8, 10, 11} are also homometric. Their interval content is equal to
ic(M5 X) = 233231.

Solution 3.1
By starting the cycle of fifths on F (x = 5) and retaining the first k
fifths, we obtain the desired scales. For some values of k, the scale
is not well formed. For instance, for k = 4, the scale 0, 2, 5, 7 is not
WF because they are two notes between 5 and 0 and three notes
between 0 and 7. The scales for k = 6, 8, 9, 10, 11 are also not WF.
250 A Compendium of Musical Mathematics

Using Douthett’s function, we compute the maximally even scales.

The results are presented in the following tables:
k WFS k Douthett J-sets
1 {5} 2 {3, 9}
2 {0, 5} 3 {2, 6, 10}
3 {0, 5, 7} 4 {1, 4, 7, 10}
5 {0, 2, 5, 7, 9} 5 {1, 3, 6, 8, 11}
7 {0, 2, 4, 5, 7, 9, 11} 6 {1, 3, 5, 7, 9, 11}
12 {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11} 7 {1, 2, 4, 6, 7, 9, 11}

Solution 3.2
By computing the generic and specific intervals, we get
X, generic Specific Y, generic Specific Z, generic Specific
1 1, 2, 3 1 1, 2 1 1, 2
2 3, 4 2 3, 4 2 3, 4
3 4, 5, 6 3 4, 5, 6 3 5, 6
4 6, 7, 8 4 6, 7, 8 4 6, 7
5 8, 9 5 8, 9 5 8, 9
6 9, 10, 11 6 10, 11 6 10, 11

Only the descending melodic minor scale Z is maximally even

and has Myhill’s property since it has two specific intervals for every
generic interval.
Solution 3.3
The tables of the generic and specific intervals are as follows:
X, generic Specific
1 1, 2
2 3 Y, generic Specific
3 4, 5 1 2, 3, 4
4 6 2 5, 6, 7
5 7, 8 3 8, 9, 10
6 9
7 10, 11
Since the scales have exactly two specific intervals for each generic
interval, X and Y are maximally even. But they do not have Myhill’s
property.
 Solutions to Exercises 251

Solution 3.4
Transpositions of X lead to only one pattern M2-M2. Thus, the car-
dinality (of X = 3) is not equal to variety (= 1). Moreover, on the
circle of fifths (0, 2, 4, 6, 8, 10, 0), the structure 1-1-4 is different of the
multiplicity.
Solution 3.5
The transpositions of X are {0, 2, 4}, {2, 4, 7}, {4, 7, 9}, {7, 9, 0} and
{9, 0, 2}: one has structure 2-2-8, two have structure 2-3-7 and the
two others have structure 3-2-7. They are 3 patterns for 3 elements,
thus cardinality equals variety. On the circle of fifths (0, 7, 2, 9, 4, 0),
the structure 2-2-1 is equal to the multiplicity.
Solution 3.6
Using the generator m, we get 0, 8, 3, 11, 6, 1, 9, 4, 12, 7, 2, 10, 5. The
scale X is the 8 first values

X = {0, 1, 3, 4, 6, 8, 9, 11}

Since the interval vector iv(X)(8) = card(X ∩ T8 (X)) = 7 =

card(X) − 1, X is a PT set by Theorem 3.1. Moreover, since X
has no three consecutive and its black keys X c = {2, 5, 7, 10, 12}
have no two consecutive notes, X is a diatonic scale. This can
been seen using u = max(−r, r) with r = N − m = 5, and
v = min(u, max(5, 10), max(−5, −10)) = 8 verifying u ≤ k ≤ v.
Solution 3.7
Taking the first nine notes in the series {mx mod N, x = 0, . . . , k −
1}, we get X = {0, 2, 4, 6, 8, 9, 11, 13, 15}. Since card(X ∩ T9 (X)) =
8 = card(X) − 1, X is a PT set. Moreover, X c = {1, 3, 5, 7, 10, 12, 14}
has no two consecutive notes. Thus, X is a diatonic scale. Since it
has one generalized tritone {0, 8}, X is a generalized major scale.

Solution 4.1

(1) Denote xM the major chord rooted on x, and xm the minor chord
rooted on x, we have

Sj (xM ) = (x + j)M , Sj (xm ) = (x − j)m

Wj (xM ) = (x + j)m , Wj (xm ) = (x − j)M
252 A Compendium of Musical Mathematics

Thus by definition of P , L, R, we have

P (xM ) = xm = W0 (xM )
L(xM ) = (x + 4)m = W4 (xM )
R(xM ) = (x + 9)m = W9 (xM )

(2) For the major triad, we have

Si Sj (xM ) = Si (x + j)M = (x + i + j)M = Si+j (xM )

Si Wj (xM ) = Si (x + j)m = (x + j − i)m = Wj−i (xM )
Wi Sj (xM ) = Wi (x + j)M = (x + j + i)m = Wj+i (xM )
Wi Wj (xM ) = Wi (x + j)m = (x + j − i)M = Sj−i (xM )

and same relations for minor triad.

Solution 4.2
Let us denote the set

{CM , EM , AbM , Cm , Em , Abm }

by {0M , 4M , 8M , 0m , 4m , 8m }. The actions are as follows:

0M 4M 8M 0m 4m 8m
T0 0M 4M 8M 0m 4m 8m
T4 4M 8M 0M 4m 8m 0m
T8 8M 0M 4M 8m 0m 4m
I3 8m 4m 0m 8M 4M 0M
I7 0m 8m 4m 0M 8M 4M
I11 4m 0m 8m 4M 0M 8M

0M 4M 8M 0m 4m 8m
S0 0M 4M 8M 0m 4m 8m
S4 4M 8M 0M 8m 0m 4m
S8 8M 0M 4M 4m 8m 0m
W0 0m 4m 8m 0M 4M 8M
W4 4m 8m 0m 8M 0M 4M
W8 8m 0m 4m 4M 8M 0M
 Solutions to Exercises 253

Solution 4.3
S is clearly the permutation associated with S1 , and W the permu-
tation of W0 . Thus, Wn = P Sn = W0 Sn = W S n . (2) P = W0 = W ,
L = W4 = W0 S4 = W S 4 , R = W9 = W S 9 . Consequently, since
Wi Wj (X) = Sj−i (X), we have
(RL)3 R = (W9 W4 )3 W9 = S73 W9 = S21 W9 = S9 W9 = W0 = P
using Si Sj (X) = Si+j (X) we see that Sn3 = S3n (mod 12 on the
indices).
Solution 4.4

DM
LP L / Bbm DM
J / Bbm

PL PL JQJ JQ
   
F m / Dm F m / Dm
LP L QJ

Solution 4.5
We have for the major triad
U V (xM ) = U (xm ) = (x + 1)m
V U (xM ) = V (x + 1)M = (x + 1)m
and for the minor triad
U V (xm ) = U (x + 6)M = (x + 7)M
V U (xm ) = V (x + 1)m = (x + 7)M
Therefore, U V = V U for all triads.
The permutations are
U = (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12)
(13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24)
V = (1, 13, 7, 19)(2, 14, 8, 20)(3, 15, 9, 21)
(4, 6, 10, 22)(5, 17, 11, 23)(6, 18, 12, 24)
The computation leads to
U V = V U = (1, 14, 9, 22, 5, 18)(2, 15, 10, 23, 6, 19)
(3, 16, 11, 24, 7, 20)(4, 17, 12, 13, 8, 21)
254 A Compendium of Musical Mathematics

Solution 4.6
Verifying the relations, we get u4 (xM ) = xM and u4 (xm ) = xm .
Furthermore,

v 2 (xM ) = v(3 − x)m = xM

u2 (xM ) = u(x + 6)M = xM

thus u2 (xM ) = v 2 (xM ). The same holds for minor triads u2 (xm ) =
v 2 (xm ). The last relation is also verified

uvu(xM ) = uv(x + 6)M = u(−3 − x)m = (3 − x)m = v(xm )

uvu(xm ) = uv(x + 6)m = u(9 − x)M = (3 − x)M = v(xM )

By noticing that u = T6 and v = I3 , the relations are immediately

verified. Using Ii Ij = Ii−j , Ti Ij = Ii+j , we have I32 = T0 = T62
(v 2 = 1 = u2 ) and T6 I3 T6 = T6 I−3 = I3 (uvu = v).
Solution 5.1
We have

E(4, 7) = aE(4, 3) = abE(1, 3)abaE(1, 2) = abaaE(1, 1)

= abaa(1) = aba(2) = ab(3) = a(0111) = 1222

Solution 5.2
We have

12222 = a(01111) = ab(4) = aba(3) = abaa(2)

= abaaa(1) = abaaaE(1, 1)
= abaaE(1, 2) = abaE(1, 3) = abE(1, 4) = aE(5, 4) = E(5, 9)

Solution 5.3
We have

E(n, 2n + 1) = aE(n, n + 1) = aaE(n, 1) = aabE(n − 1, 1)

= aabn−1 E(1, 1)
= aabn − 1(1) = aa(0n−1 1) = a(1n−1 2) = (2n−1 3)

Symbol 2n−1 is 2 repeated (n − 1) times. The first Aksak rhythms

are E(2, 5) = (23), E(3, 7) = (223), E(4, 9) = (2223), E(5, 11) =
 Solutions to Exercises 255

(22223) etc. Some Aksak rhythms do not belong to this hierarchy:

E(4, 11) = 3332, E(5, 13) = 32323, E(7, 17) = 3232322. But not all
Aksak rhythms are Euclidean. Aksak rhythms were widely used by
the Turkish composer Ahmet Adnan Saygun (1907–1991) in many
piano pieces.

Solution 5.4
By computing the distances and occurrences, we see that the rhythm
E(4, 9) = 2223 which is {0, 2, 4, 6, }9 is E-deep:

0 2 4 6
0 0 2 4 3 d 1 2 3 4
2 2 0 2 4
i 0 3 1 2
4 4 2 0 2
6 3 4 2 0

Since the rhythm 2223 is the rotation {2j mod 9, j = 0, 1, 2, 3}

and by applying Theorem 5.11, we see that this rhythm is
E-deep. Furthermore, the hierarchy of Aksak rhythms {0, 2, 4, . . . ,
2(n−1)}2n+1 is the rotation {2j mod n, j = 0, 1, . . . , n−1}. Applying
Theorem 5.11 shows that E(n, 2n + 1) are E-deep.

Solution 5.5
The fume-fume rhythm is isomorphic to the pentatonic scale
0-7-2-9-4. It is well formed and maximally even since they are two
c-distances for each generic interval:

0 2 4 7 9
d 1 2 3 4 5 6
0 0 2 4 5 3
i 0 3 2 1 4 0
2 2 0 2 5 5
4 4 2 0 3 5
c 1 2 3 4
7 5 5 3 0 2
d 2, 3 4, 5 7, 8 9, 10
9 3 5 5 2 0

Solution 5.6
(1) The height of ω = 3232323 is even and no cyclic shift of ω can
be factorized in two words uv such that h(u) = h(v). (2) The length
of ω is 2 + 1 with  = 3. The height of the prefixes of length 3 (323,
332, 233, 232) is 8 or 7 for 232. (3) Let u = 323 and v = 2323, the
256 A Compendium of Musical Mathematics

height h(v) = h(u) + 2. Furthermore,

(u, v) = (323, 2323) = s(323, 323) = sr(23, 23)

= srs(3, 23) = srss(3, 3) = srssr(1, 1)

Thus, α = srssr, with |α|s = 3 odd.

Solution 5.7
Let u = 3322, v = 32322, such that h(v) = 12 = h(u) + 2. We have

(u, v) = (3322, 32322) = r(322, 2322) = rs(322, 322) = rsr(22, 22)

= rsrss(2, 2) = rsrsss(1, 2) = rsrssss(1, 1)

Thus, α = rsrssss, with |α|s = 5 odd.

Solution 5.8
The Dick path shows that ω = C(8, 3) = aaabaaabaab and the stan-
dard factorization is (aaab, aaabaab). The determinant
   
|ω1 |a |ω2 |a 35
det = det =1
|ω1 |b |ω2 |b 12

Thus, the matrix belongs to SL2 (Z). The standard factorization with
ω1 = C(3, 1) = aaab = aub with u = aa and ω2 = C(5, 2) =
aaabaab = avb with v = aabaa can be rewritten as the palin-
dromic factorization ω = (ava, bab). Considering the multiples of
3 (or 8 reading from left to right) (3, 6, 9, 1, 4, 7, 10, 2, 5, 8, 0), we
get the black keys ω1 = 3, 6, 9, 1 = {1, 3, 6, 9} and the white keys
ω2 = 4, 7, 10, 2, 5, 8, 0 which is the diatonic scale:

ω2 = {0, 2, 4, 5, 7, 8, 10} = dia811

Solution 6.1
We have

A(x) = (1 + x5 + x8 + x13 ) = (1 + x5 )(1 + x8 )

= (1 + x)(1 − x + x2 − x3 + x4 )(1 + x8 ) = Φ2 (x)Φ10 (x)Φ16 (x)

Since A(1) = 4 = A and Φ16 | A(x), conditions (T1 ) and (T2 ) are
satisfied. A tiles Z16 by CM theorem.
 Solutions to Exercises 257

Solution 6.2
(1) Let us set
xN − 1
QN (x) = = 1 + x + x2 + · · · + xN −1
x−1
We have A(x)B(x) = QN (x), and C(x) = A(x)Qk (xN ). Thus,
C(x)B(x) = A(x)B(x)Qk (xN ) = QN (x)Qk (xN ) = QkN (x)
(2) This means that the cyclotomic factors of C are those of A, and
the cyclotomic polynomial Φd is such that d | kN and d  N . Let
pα ∈ SA and q β ∈ SQkN , then since A tiles ZN p | N implies that
pα q β | kN . Thus, Φpα qβ | QN | C.
Solution 6.3
Clearly, the shape of A: 11 11 11 shows that A does not tile. More-
over,
A(x) = 1 + x + x3 + x4 + x6 + x7 = (1 + x)(1 + x3 + x6 ) = Φ2 (x)Φ9 (x)
A verifies (T1 ) since SA = {2, 32 } and A(1) = 6 = 2 × 3 = |A|. But
Φ2×9 = Φ18 does not divide A(x).
Solution 6.4
(1) Since B(x) = xn A(x), A(1) = B(1) = |A| = |B|. (2) For
every cyclotomic polynomial Φd (x), we have Φd (x) | A(x) ⇔
Φd (x) | B(x).
Solution 6.5
(1) We have B(x) = A(xn ), and (T1 ) follows since A(1) = B(1).
(2) Let SA = {q1α1 , . . . , qm
αm } be powers of distinct primes. From the

property,

p Φps (x) if p is a factor of s
Φs (x ) = (9.4)
Φs (x)Φps (x) if p is not a factor of s
we see that if s ∈ SB = SpA , then s = pqiαi if p is a factor of qiαi or
s = qiαi if p is not factor of qiαi . Consequently,
SB = {pα+1 , pα ∈ SA } ∪ {q β ∈ SA , q prime = p}
and A satisfies (T2 ) iff B satisfies (T2 ). The result is also true for
any n. The proof consists in iterating over each prime resulting from
the decomposition of n.
258 A Compendium of Musical Mathematics

Solution 7.1
Let’s label the circle clockwise from 0 to 5. The generalized tritone
is (0, 3). We write the chord diagrams by connecting the points of
3 elementary intervals. There exists 5 chord diagrams whose Gauss
words are as follows:

D1 (05)(12)(34) 112233 aaa

D2 (01)(25)(34) 112332 aca
D3 (01)(24)(35) 112323 abb
D4 (03)(15)(24) 123132 bbc
D5 (03)(14)(25) 123123 ccc

The subseries 450123 belongs to diagram D4 . The all-interval

series 150324 has chord diagram D4 . The all-interval series has a
chord diagram that has two 3-step intervals. The only diagram with
2 consecutive strings a is the D4 diagram.

Solution 7.2
The group G has 24 elements

1 b b2 b3 b4 b5
b6 b−1 b−2 b−3 b−4 b−5
a ab ab2 ab3 ab4 ab5
c cb cb2 cb3 cb4 cb5

The action of a on the series a(S) = 8391t0e564728 does not

change the chord diagram.

Solution 7.3
The series S = 03e4128t5976 is all-interval series. Moreover, the set
Ac = {5, 6, 7, 8, 9, t} of its second hexachord is the transposition a
tritone up (or down) T6 (Ac ) = A = {0, 1, 2, 3, 4, e} of its first hexa-
chord. Since T6 (A) = Ac , S ∈ T is a transpose tone row, and since
T9 (−A) = Ac , S ∈ R is a reverse series.

Solution 7.4
The series S is not an all-interval series. Furthermore, by listing the
transpositions and inversions of the first hexachord, we see that S ∈ /
T ∪ R ∪ T  ∪ R .
 Solutions to Exercises 259

Solution 7.5
We have
σ(S) = 901738425BA6 D111 σ 5 (S) = 34179A065B82 D10
σ 2 (S) = 69712038B54A D36 σ 6 (S) = 2371649AB508 D105
σ 3 (S) = A61789205B34 D105 σ 7 (S) = 8217A3645B90 D36
σ 4 (S) = 4A710689B523 D10 σ 8 (S) = 087142A3B569 D111

The serial plan of the piece is as follows:

2 3 4
& & &
1 5
+ D111e D36 e D105e D10 n
8 7 6

Solution 7.6
We have
 
−1 0123456789t e
B =
2te057614983

and
 
−1 0123456789t e
AB =
e9t304128657
= (0e72t54)(196)

Thus, the Cayley distance is 2 (cycles).

Solution 8.1
Compute all differences (written multiplicatively):

1 a a3 b a2 b2
1 1 a6 a4 b2 a3 b
a a 1 a5 ab2 a4 b
a3 a3 a2 1 a3 b2 a6 b
b b ba6 ba4 1 ba3 b
a2 b2 a2 b2 a2 b2 a6 a2 b2 a4 a2 b4 a2 b2 a3 b
260 A Compendium of Musical Mathematics

Using the group relations, we get

a3 b = a(a2 b) = a(ba) = a8 (ba) = a6 (a2 b)a = a6 (ba)a = a4 (a2 b)a2
= a4 (ba)a2 = a2 (a2 b)a3 = a2 (ba)a3 = (a2 b)a4 = ba5
For instance,
a2 b2 a3 b = a2 b2 ba5 = a2 a5 = 1
Solution 8.2
The multiplication xy −1 leads to the following table:
a a2 a4 b b2
a 1 a6 a4 ab2 ab
a2 a 1 a5 a2 b2 a2 b
a4 a3 a2 1 a4 b2 a4 b
b ba6 ba5 ba3 1 b2
b2 b2 a6 b2 a5 b2 a3 b 1
Solution 8.3
The addition a − b leads to the following table:
1 3 13 16 17
1 0 2 12 15 16
3 19 0 10 13 14
13 9 11 0 3 4
16 6 8 18 0 1
17 5 7 17 20 0
The multiplier p is a prime divisor of n = k − λ = 4, thus p = 2.
On has
2D = {2, 6, 5, 11, 13} = T10 (D)
Solution 8.4
The multiplier is p = 2, since
2D = {2, 8, 10, 20, 1} = T19 (D)
Solution 8.5
D has parameters (31, 6, 1). Applying the first multiplier theorem,
we get p = k − λ = 5.
 Solutions to Exercises 261

Solution 8.6
Applying the BRC theorem, we consider the equation z 2 = 10x2 −
2y 2 . It has no solution, since −2 = 8 mod 10. The design (67, 12, 2)
does not exist.
Solution 8.7
For instance, Tom Johnson finds the graphic representations (see
Fig. 9.1).

Fig. 9.1. Block design (7, 3, 2) by Tom Johnson.

262 A Compendium of Musical Mathematics

Solution 9.1
The first convergents are 1/2, 1/4, 5/19, 111/422, 449/1707, etc.
 6 19
5 ≈ 31.948 and 25 = 32. This means that after 19 minor thirds,
we are in the vicinity of 5 octaves. The difference is

25 519
c= = , and 1200 log 2 (c) = 2.8 cents
(6/5)19 214 319

Solution 9.2
The first convergents are 1/3, 9/28, 19/59, 47/146, 207/643, etc.
 5 28
4 ≈ 516.988 and 29 = 512. This means that after 28 major
thirds, we are in the vicinity of 9 octaves. The difference is

(5/4)28 528
c= = , and 1200 log 2 (c) = 16.8 cents
29 265
The next approximation 19/59 is better, and the comma is

219 2137
c= 59
= 59 , 1200 log 2 (c) = 7.5 cents
(5/4) 5

Solution 9.3

Ratios Cents 3/2 702

1 0 8/5 814
7/6 267 5/3 884
6/5 316 7/4 969
5/4 386 9/5 1018
4/3 498 11/6 1049
7/5 583 2 1200

Solution 9.4
From the graph

5 3×5 32 × 5 33 × 5 34 × 5

1 3 32 33 34
 Solutions to Exercises 263

we compute the entries of the following table:

Ratios Cents 45/32 590

1 0 3/2 702
135/128 92 405/256 794
9/8 204 27/16 906
5/4 386 15/8 1088
81/64 408 2 1200

Solution 9.5
#A Structure a b a (cents) b (cents)
3 aba 5/4 32/25 386 427
7 abababa 128/125 625/512 41 345
13 a ba3 ba3 ba2
2 128/125 510 /223 41 263
19 a3 ba5 ba5 ba3 128/125 516 /237 41 181
25 a4 ba7 ba7 ba4 128/125 522 /251 41 99
31 a5 ba9 ba9 ba5 128/125 528 /265 41 17
Solution 9.6
#A Structure a b a (cents) b (cents)
3 aba 6/5 25/18 316 569
7 abababa 125/108 648/625 253 63
11 aba2 baba2 ba 648/625 57 /25 37 63 191
19 a2 ba4 ba3 ba4 ba2 648/625 515 /211 315 63 65
23 a2 ba5 ba5 ba5 ba2 648/625 519 /214 319 63 3
Solution 9.7
The tuning is [1, 9/8, 5/4, 4/3, 3/2, 5/3, 9/5, 2]. The simplest solu-
tions of equation 7x + 11y + 16z = 0 are |−3, −1, 2 associated with
25/24 and |−7, 3, 1 associated with 135/128.
Solution 9.8
The tuning is [1, 8/7, 7/6, 4/3, 3/2, 12/7, 7/4, 2]. The kets |6, −2, −1
with comma 64/63 (27 cents) and |1, 3, −2 with comma 54/49
(168 cents) are solutions of 7x + 11y + 20z = 0.
This page intentionally left blank
 Index

A Bouchet, André, 119

Aaron, Pietro, 218, 229 Boulez, Pierre, 184–185, 191
affine group, 23, 26 braid group, 129
Agmon, Ethan, 65 Brăiloiu, Constantin, 119
all-combinatorial, 186 Brjuno function, 228
Amiot, Emmanuel, xiv, 31, 47, 151 Broué, Michel, 65
Andreatta, Moreno, 98 Burnside lemma, 20
aperiodic monoid, 113 Busoni, Feruccio, 1
apotome, 216
Armstrong, Drew, 141 C
Aron, Simha, 119 Cannas, Sonia, 98
Artin group, 129 canons, 151
associahedra, 141 Carey, Norman, xii, 60
atonal triad, 96 Carter, Elliott, 183
automorphisms, 79, 82 Catalan numbers, 141
centralizer, 94
B cents, 214
Baaren, Kees van, 192 Chemillier, Marc, 113, 120
Babbitt hexachord theorem, 43 Childs, Adrian, 97
Babbitt, Milton, xiii, 183, 185, 192 Chopin, Frédéric, 98
Bach, Jean-Sébastien, 151 chord diagrams, 177, 183
Balzano, Gerald J., 65 Christoffel words, 125, 130
Barbour, James Murray, 217 chromatic distance, 60
Barraqué, Jean, 182, 184, 188–189 Clampitt, David, xii, 60, 109, 127
beats, 214 Clough, John, 116
Berg, Alban, 151, 178, 190 Cohn, Richard, xii, 77
Bernstein, Leonard, 183 Colbourn, Charles J., 197
Besada, José Luis, 114 collections, 24
billiard words, 125 Coltrane, John, 102
block designs, 199 combination product sets, 236

265
266 A Compendium of Musical Mathematics

combinatorial designs, 197 equal division of the octave,

combinatoriality, 185 214
comma Erdös, Paul, 117
Hellgouarch comma, 237 Estrada, Julio, xi, 16
Holderian comma, 215 Euclidean rhythms, 115
Pythagorean comma, 215, 218 Euler, Leonhard, 230
septimal comma, 216 Euler totient function, 25, 232
syntonic comma, 215 Euler–Fokker Genera, 230
common tone theorem, 39 evaluation (of a word), 135
complexity (of a word), 123 Evans, Bill, 101
concatenation, 109
concert A, 213 F
conjugated modes, 128 factor, 110
conjugation, 18 Farey, John, 231
continued fractions, 220 Farey sequences, 231
convergents, 220 Farey tunings, 231
cosets, 18 Fauré, Gabriel, 151
Costère, Edmond, xi, 1, 16, 183, 185 Fechner law, 214
Coven, Ethan, 164, 183 Fibonacci words, 122
Crumb, George, 96 Fiore, Thomas, 94
cyclic group, 23–24, 180 Fisher’s inequality, 201
cyclic tunings, 235 fixed points, 17
cyclotomic polynomials, 154 Fokker, Adriaan, 231
Ford circles, 233
D Forte, Allen, xi, 1, 16, 53
Dallapicolla, Luigi, 184, 192 Fourier transform, 47
de Bruijn, Nicolaas Govert, xiii, 163 frequency ratio, 2
Debussy, Claude, 7 Fripertinger, Harald, 17, 179
deep rhythm, 117 Fuglede, Bengt, 165
derived Catalan numbers, 142
diatonic distance, 60 G
diatonic scales, 65, 68, 72 Gamer, Carlton, 117, 197
diesis, 216 generalized interval system, 31
difference sets, 197 generating power, 191
dihedral group, 23, 25, 180 Goeyvaerts, Karel, 182, 185
Dinitz, Jeffrey H., 197 Gollin, Edward, 97
dodecaphonic knots, 177 grammar, 113
Douthett, Jack, 58, 61, 98, 116, 227 Grisey, Gérard, xiii
dual group, 94, 152 group action, 17
Dufourt, Hugues, xiii group factorization, 152
Dyck paths, 142
Dyck words, 121 H
Hába, Alois, 1
E Hadamard matrix, 166
Eimert, Herbert, 182 Hajós groups, xiii, 163
Ellis, John, 115 Hanson, Howard, xi, 16
 Index 267

harmonic complexity, 219 Klein group, 185

harmonic gap, 219 Klumpenhouwer, Henry, xii
harmonic metrics, 218 Klumpenhouwer networks, 77
Hauer, Josef Matthias, xiii, 1 Knuth congruences, 132
height (of a word), 111 Knuth, Donald, xiii
Hellegouarch, Yves, 218, 237 Krenek, Ernst, 185
heptatonic modes, 139 Kreweras numbers, 144
Hertz, 213
Herzog, Eduard, 182 L
hexatonic groups, 92 Lagrange theorem, 19
hexatonic modes, 139 Lanzarotto, Gretta, 171
Hindemith, Paul, 151 Lateef, Yusef, 1
Hölder, William, 229 leimma, 216
homometric multiplets, 54 Lenot, Jacques, 185, 192
homometric sets, 48 Lerdhal, Fred, 1, 113
Huygens, Christiaan, 231 Lévy, Fabien, 167
Lewin, David, xi, 31, 54, 77
I lexcicographic order, 110
Ilomäki, Tuukka, xiii, 187 Liang, Frank, 234
incidence matrix, 201 limited transposition sets, 5
injection number, 38, 58 Lindermayer system, 113
interval class, 2 Loquin, Anatol, xi, 1
interval class vector, 37 Louvier, Alain, 57
interval function, 35 Lumma, Carl, 237
interval sum, 44 Lyndon words, 114
interval vector, 37
inversion, 3 M
isographic networks, 78
major scales, 72–73
Manoury, Philippe, 113
J
mask polynomials, 49, 153
Jankó, Paul von, 229 maximally even (rhythms), 116
Jelinek, Hanns, 183 maximally even (sets), 61
Johnson, Tom, 117, 122, 160, 197, Mazzola, Guerino, 16
203, 205, 209 mean deviation, 217
JQZ group, 86 meantone temperament, 229
Messiaen modes, 7, 10
K Messiaen, Olivier, 7, 97, 151
Kassel, Adrian, 237 Meyerowitz, Aaron, 164
Kassel, Christian, 237 minor scales, 73
Kerkez, Boris, 98 Möbius function, 115
Kirkman numbers, 144 monoid, 110
Kirkman, Thomas Penyngton, 206 Morris, Robert, 3, 179, 190
Kirkman’s ladies, 206 multiplier, 198
Kleene star, 112 Murail, Tristan, xiii
Klein, Fritz Heinrich, 183 Mutterakkord, 183
268 A Compendium of Musical Mathematics

N R
n-limit, 233 Rameau groups, 104
Narayana numbers, 143 Rameau–Schillinger operators, 99
negative isographies, 78 rational languages, 112
Netto’s theorem, 208 recognizable languages, 112
Noll, Thomas, xiii, 94, 109, 127 Reiner, David, 179
Nono, Luigi, 184, 190 resolvable designs, 204
rhythmic oddity, 119
O Riemann, Hugo, xii, 77
octatonic groups, 93 Riotte, André, 182
octatonic modes, 140 Rosenblatt, Joseph, 51
orbits, 17 Roux, Sébastien, 167
overtones, 213 RSK correspondence, 134
Russel, Georges, 1
P
Papadopoulos, Athanase, xiv, 213 S
parallel class, 204 S/W group, 86
Partch, Harry, 234 Sands, Arthur, xiii, 163
partials, 213 Savart, Félix, 214
Parzysz, Bernard, 10, 183 Saygun, Ahmed, 119
Patterson function, 49 scale, 3
pentatonic modes, 137 Schat, Peter, 185
perfect tilings, 160 Schensted, Craige, 133
periodic words, 123 Schönberg, Arnold, xiii, 151, 184, 187,
Pernazza, Ludovico, 171 192, 210
Piňos, Aloı̈s, xi, 16 Schützenberger, Marcel-Paul, 133
pitch class set, 2 semi-combinatorial, 185
pitch classes, 2 semiconvergents, 221
plactic modal class, 136 serial groups, 192
plactic monoid, 132
set class, 3
PLR group, 86
similarity measures, 187
Pólya’s theorem, 12, 21
Skalkottas, Nikos, 183
Popoff, Alexandre, 95
Slonimsky, Nicolas, 1
Posadas, Alberto, 114
Sós, Vera T., 234
positive isographies, 78
special Sturmian monoid, 129
primitive word, 110
spectral sets, 165
progressive transposition sets, 58
spectral unit, 52
Pythagoras, 229
stabilizer, 17
Pythagorean temperaments, 235, 239
star-free languages, 112
Pythagorean tunings, 228
Starr, Daniel, 16
Steedman, Marc, 113
Q
Steiner systems, 199
quarter tones, 10 Steiner triple systems, 199, 206
 Index 269

Steinhaus conjecture, 234 tuning systems, 215

Stevin, Simon, 229 twelve-tone rows, 177
Stockhausen, Karlheinz, 190
Stravinsky, Igor, 151 U
Sturmian monoid, 127 uniform triadic transformations, 93
Sturmian morphisms, 127
Sturmian word, 123 V
superpartial ratios, 230
Surányi, János, 234 Vieru, Anatol, 1
symmetric designs, 199 Vlijmen, Jan van, 185
symmetric group, 23 Vuza, Canons, xiii, 167, 169
syntactic monoid, 112 Vuza, Dan Tudor, 167
Szabo, Sandor, xiii
W
T W-deep, 117
T/I group, 78 Wagner, Richard, 98
T/M group, 81 Webern, Anton, 151, 182–183,
Taslakian, Perouz, 118 190–191
three-gap theorem, 234 well-formed scales, 60
Thue–Morse word, 122 Wilson, Ervin, 235–236
tile, 152 Wyschnegradsky, Ivan, 57, 232
tiling pairs, 152
transformations J,Q,Z, 87 Z
transformations P ,L,R, 77, 84 Z-relation, 48
transitive actions, 17 Zarlinian temperament, 231
transposition, 2 Zimmermann, Bernd Alois, 184, 191