Linear Contextual Transformations

Rachel Wells Hall

June 11, 2009

Abstract
This paper considers groups of musical contextual transformations,
the most famous of which is the group of neo-Riemannian transforma-
tions acting on the set of major and minor triads. David Lewin (1982,
1987) rst described neo-Riemannian transformations, building on work
by von Oettingen (1866) and Hugo Riemann (1880). In general, contextual
transformations act on multisets of pitches or pitch classes and commute
with transposition and inversion. Music theorists have studied contextual
transformation groups extensively; in particular, Lewin (1987), Kochavi
(1998), Cohn (1997), and Fiore and Satyendra (2005) used discrete group
theory, while Clough (1998) and Gollin (1998) modeled contextual trans-
formations with symmetries of discrete geometric spaces. We represent
contextual transformations as linear transformations acting on sequences
of pitches or pitch classes belonging to the continuous geometric voice-
leading spaces developed by Callender, Quinn, and Tymoczko (2008).
Subtleties arise that have not been explored in discrete models. Con-
nections to tuning theory, neo-Riemannian theory, serial transformations,
and musical examples from di Lasso, Brahms, and Schubert demonstrate
the benets of this approach.
1 Introduction
The neo-Riemannian transformations, formalized by David Lewin [12, 11] fol-
lowing work by the music theorists Arthur von Oettingen [14] and Hugo Rie-
mann [15], constitute a group that acts on the set of consonant triads.
1
This
group is generated by three involutions that are bijections between the major
triads and the minor triads: the relative transformation (R), which associates a
major triad with its relative minor and vice versa; the parallel transformation
(P), which associates parallel major and minor triads; and the leittonwechsel
transformation (L), which associates a major triad with its mediant and a mi-
nor triad with its submediant (for example, L associates C major and E minor).

Department of Mathematics, Saint Josephs University, 5600 City Avenue, Philadelphia,

PA 19131, USA. [email protected]
1
Consonant triads refers to major and minor triads. Henceforth, I use triads to denote
consonant triads.
1
&

C a

b
C c

C e

C0 a1

b
C0 c0

C0 e2
R P L R
P
L
Figure 1: Generators of the group of neo-Riemannian transformations. Capital
and lowercase letters indicate major and minor triads, respectively. On the
left, all triads are in root position; on the right, they are arranged so as to
create smooth transitions (voice leadings) between their notes. The numbers 0,
1, and 2 indicate the registration of the chord: they represent root position, rst
inversion, and second inversion, respectively. (Root position means the lowest
note gives its name to the triad, the middle note is a third above it, and the top
note is a fth above the root, rst inversion has the third as its lowest note, and
second inversion has the fth as its lowest note.)
Figure 1 depicts examples of these progressions, which are ubiquitous in tonal
music, from the Renaissance to the present day.
Neo-Riemannian transformations are examples of a larger class of transfor-
mations acting on major and minor triads. The most recognizable transfor-
mation is transposition (T
k
)shifting pitches upward by some real number of
semitones k. (For example, the transposition T
2
shifts pitches upward by two
semitones, transposing a C major triad to a D major triad.) Inversion (I) is an
involution that reects every pitch around some xed origin such as middle
C (this means that the inversion of a C major triad is an F minor triad and
vice versa, as in gure 2). Like P, L, and R, inversion is a bijection between
the major and the minor triads. However, inversion does not commute with
transposition (in fact, IT
k
= T
k
I). For example, inversion maps a C major
triad to a F minor triad but maps D major to E minor. Inversions are more
dicult to identify than transpositions and neo-Riemannian transformations:
the inversion from C major to F minor is not normally considered the same
chord progression as the inversion from D major to E minor. In contrast, the
relationship between C major and its relative minor, A minoralso its inver-
sion in Gis commonly identied with the relationship between D major and
its relative minor, B minor, which is its inversion in A. The relative transfor-
mation captures this similarity: the axis of inversion follows a chord as it is
transposed (in other words, R commutes with transposition).
Neo-Riemannian transformations belong to a larger class of so-called contex-
tual transformationscontextual because their action depends on the objects
they act on. In the case of the relative transformation, the axis of inversion is
determined by the chord being inverted. Mathematically, contextual transfor-
mations commute with both transposition and inversion (though not necessarily
with each other). In addition to the neo-Riemannian transformations, music the-
orists have dened contextual transformations that apply to non-triadic chords
2
and ordered sequences of pitch classes [12, 3, 5, 10]. Contextual transformations
appear in both tonal and atonal music.
Music theorists have long employed geometric models to describe relation-
ships between chords and scales (the circle of fths, Heinichens circle of major
and minor triads) or to depict tuning systems (Eulers speculum musicum [4], the
Oettingen/Riemann tonnetz [14, 15]). However, a unied geometrical perspec-
tive emerged only recently in the work of Callender, Quinn, and Tymoczko [16, 1]
(henceforth, CQT). CQT recognized that basic equivalence relations such as oc-
tave identication may be realized as symmetry relations on R
n
. This allowed
them to dene a family of quotient spaces that subsume many of the geomet-
ric models proposed in the literature. Geometrical music theory applies CQT
models to musical analysis.
Although Clough [2] and Gollin [7] modeled contextual transformations as
symmetries of discrete geometric spaces, contextual transformations have not
yet been incorporated into geometrical music theory. This paper represents con-
textual transformations as groups of ane linear transformations acting on the
CQT spaces of ordered sequences of pitches or pitch classes, thus connecting
prior work in music theory with the rich mathematical theory of linear trans-
formations. Moreover, because the geometry of linear transformations is well
understood, matrix groups provide a bridge between geometrical music theory
and the more traditional algebraic theory of contextual transformations. Af-
ter a brief introduction to geometrical music theory (section 2), we consider
matrix representations of contextual transformation groups (section 3). Sec-
tion 4 discusses the action of these groups (in particular, the implications of
equal temperament). Finally, section 5 considers musical examples, including
the neo-Riemannian transformations and their matrix representations.
2 Geometrical music theory
The basics of geometrical music theory are as follows. Pitch is frequency mea-
sured on a logarithmic scale, with twelve units of pitch (semitones) to an octave.
We arbitrarily assign 440 Hzthe A above middle Cto pitch 9.
2
This pitch
is also labeled A4, with the 4 indicating the octave. Middle C is pitch 0, also
labeled C4. Pitches lie on a continuum; integer pitches form twelve-tone equal
temperament (12-tet). Pitch vectorsn-tuples of pitchesrepresent pitches
sounding simultaneously, with each coordinate representing an instrument or
voice.
3
The collection of n-element pitch vectors (R
n
) constitutes n-dimensional
ordered pitch space. For example, suppose we use three-dimensional ordered
pitch space to model pitches sung simultaneously by a tenor (the x-coordinate),
alto (the y-coordinate), and soprano (the z-coordinate). Then (0, 4, 7) means
that the tenor, alto, and soprano sing C4, E4, and G4, respectively, and (0, 7, 4)
means that they sing C4, G4, and E4, respectively. Our perception (and musical
practice) gives the collection of pitch vectors many symmetries: if two voices
2
Another common choice for A 440 is 69, the standard MIDI assignment.
3
Pitch vectors are called musical objects by CQT.
3
Transposition Permutation
v T
k
(v), k R v P
ij
(v)
T
k
(v) = v + k1
& w
w
w
w
w
w
#
& w
w
w
w
w
w
T
2
(0, 4, 7) = (2, 6, 9) P
23
(0, 4, 7) = (0, 7, 4)
Octave shift Inversion
v O
i,n
(v), n Z v I(v) = v
O
i,n
(v) = v + 12ne
i
& w
w
w
w
w
w
& w
w
w w
w
w
b
O
1,1
(0, 4, 7) = (12, 4, 7) I(0, 4, 7) = (0, 4, 7)
Figure 2: Representing elements of the groups O, P, T, and I by ane linear
transformations.
exchange pitches, if one voice shifts by an octave, or if all voices shift by the
same amount, the resulting vectors will represent similar sounds. For exam-
ple, (0, 4, 7), (0, 7, 4), (12, 4, 7), and (2, 6, 9) are all similar on some levelthe
rst three involve C major triads, though with dierent instrumentation and
registration, and the last, (D4, F

4, A4) is also a major triad.

4
Each of these similarity relations are dened by group actions: elements of
the permutation group (P) reorder pitch vectors; elements of the group of octave
shifts (O) move an individual pitch in a vector up or down by some number of oc-
taves; elements of the transposition group (T) shift all the pitches in a vector up
or down by a xed amount; and the nontrivial element of the inversion group (I)
reects a pitch vector in the origin. The elements of these groups are ane linear
transformations
5
acting on R
n
. Using e
i
to represent ith standard basis vector, 1
to represent (1, 1, . . . , 1), P
ij
to represent the permutation matrix that exchanges
pitches in voices i and j given by P
ij
(. . . , v
i
, . . . , v
j
, . . .) = (. . . , v
j
, . . . , v
i
, . . .),
we can represent them as in gure 2.
These groups dene equivalence relations on R
n
by identifying points in the
same orbit;
6
music theorists have given names to these orbits. The equivalence
4
Occasionally, the coordinates of a pitch vector are ordered by time, rather than instrument.
In this case, the vector represents a melodic fragment. This model is useful in considering
serial transformations, as in sections 3.4.3 and 5.6.
5
An ane linear transformation on R
n
is a transformation of the form v Mv+b,where
M is a n n matrix and b are n-element vectors, both with real entries (we assume that a
vector can be written in either row or column form, as required by the context).
6
If a group G acts on a set S, then the orbit of s in S is the set {gs : g G} of images of
4
class of a single pitch under octave shift is called its pitch class. Pitch classes
equivalence classes modulo Oare indicated by letter names (C); an ordered
sequence of pitch classes is a pitch class vector. A chord is an unordered multiset
of pitch classes (that is, an orbit in R
n
under the action of OP, the group
generated by O and P). For example, any combination of C, E, and G pitch
classes forms a C major chord. The set of major chords is an example of
a chord type (also called a transpositional set class)a set of chords related
by transposition. Finally, a set class is an orbit of chords under the action of
transposition and inversion. Major and minor triads form a set class.
Identifying (gluing together) orbits of these groups produces a family of
singular quotient spaces (also called orbifolds). For example, the set of pitch
classes constitutes the circle R/O (that is, R/12Z).
7
Pitch class vectors make up
the torus R
n
/O = T
n
. Points in R
n
/OP represent chords, points in R
n
/OPT
represent chord types, and points in R
n
/OPTI represent set classes. Cardinal-
ity equivalence (C), which identies points whose coordinates form the same
set, ignoring multiplicities, is an equivalence relation not arising from a group
action. CQT spaces are quotients of n-dimensional ordered pitch space by any
combination of the ve OPTIC equivalence relations. (See Table S2 in [1] for
a full description of CQT spaces).
2.1 Progressions and voice leadings
A sequence of pitch vectors describes both successive harmonies (indicated by
individual vectors) and melodic lines (indicated by the sequence of pitches in any
particular coordinate, or voice). In musical terms, a sequence of pitch vectors
models counterpoint. Musicians call a transition from one harmonic state to
another, indicating how the pitch in each voice changes, a voice leading.
8
An
ordered pair of pitch vectors (v, w) R
n
R
n
represents a voice leading from
v to w, denoted vw. So (0, 4, 7)(1, 4, 8) means that the rst voice moves
from C4 to B3, the second voice stays on E4, and the third voice moves from
G4 to G

4. The voice leading vw is modeled by the line segment in R

n
v + t(w v), t [0, 1].
For example, the voice leading (0, 4, 7)(1, 4, 8) corresponds to the line seg-
ment (0, 4, 7) + t(1, 0, 1) for t [0, 1].
Like pitch vectors, voice leadings can resemble each other to a greater or
lesser degree. Shifting an entire vocal line up or down an octave disrupts it less
s under the action of G.
7
In general, if a group G acts on a set S, then S/G denotes the quotient or orbit space
of S modulo the equivalence relation dened by the action of G; each element of S/G is an
orbit.
8
I depart from CQTs terminology here. They call a succession of pitch vectors a progres-
sion and use voice leading to refer to a class of progressions that are equivalent modulo
the ordering of their voices. For example, they consider (C4, E4, G4)(C4, E4, A4) and (C4,
G4, E4)(C4, A4, E4) instances of the same voice leading. Formally, if is a permutation,
then (v)(w) and vw are equivalent. Since the linear transformation model assumes
that we x some ordering of voices, I will use a more restrictive denition of voice leading.
5
than shifting individual pitches within a line up or down an octave. Consider
the three voice leadings:
(i) (C4, E4, G4)(C4, E4, A4) (ii) (C5, E4, G3)(C5, E4, A3)
(iii) (C4, E4, G4)(C4, E4, A3)
Voice leading (i) is more like (ii) than it is like (iii): in voice leadings (i) and
(ii), each voice sounds the same melody, modulo transposition of the entire
line by an octave. In (iii), the third voice leaps down ten semitones rather than
up two. However, all three voice leadings are similar in that they connect a C
major triad to an A minor triad. Vector notation clearly dierentiates these
voice leadings:
(i) (0, 4, 7)+t(0, 0, 2) (ii) (12, 4, 5)+t(0, 0, 2) (iii) (0, 4, 7)+t(0, 0, 10)
each for t [0, 1].
These notions of similarity suggests equivalence relations on voice leadings:
two voice leadings vw and xy are said to be individually octave equiva-
lent if they start and end with pitch vectors that are equivalent under octave
identication: that is, v x and w y (mod 12). An ordered pair of pitch
class vectors (v, w) is sucient to denote an individual octave equivalence class.
For example, the three voice leadings in the previous paragraph are instances
of the progression ((0, 4, 7), (0, 4, 9)). Two voice leadings are uniformly octave
equivalent if there exists an octave shift O such that x = O(v) and y = O(w).
That is, vw and xy are uniformly octave equivalent if and only if
v + t(w v) x + t(y x) (mod 12) for all t [0, 1].
Voice leadings (i) and (ii) are uniformly octave equivalent to each other and
individually octave equivalent to (iii). We can represent the uniform octave
equivalence classes of voice leadings (i), (ii), and (iii) in the following notation:
(i) (0, 4, 7) + t(0, 0, 2) (ii) (0, 4, 7) + t(0, 0, 2) (iii) (0, 4, 7) + t(0, 0, 10)
each for t [0, 1].
Geometrically, if two voice leadings, each represented by a directed line seg-
ment, are uniformly octave equivalent, then their images coincide under the
quotient map R
n
R
n
/O for all t [0, 1]. Note that mere coincidence of paths
is not sucient. For example, (0, 0)(0, 12), (0, 0)(0, 12), and (0, 0)(0, 24)
are not uniformly octave equivalent: although they traverse the same set of
points, the rst and second wind once around the torus in opposite directions
and the third winds twice. The three voice leadings are noticeably dierent
musically.
CQT represent a uniform octave equivalence class of voice leadings by two
pitch class vectors v, w T
n
and a vector d in R
n
, where d wv (mod 12).
The vector d indicates how each voice moves: the notation v
d
w means that
the ith voice ascends d
i
pitches from pitch class v
i
to w
i
. Geometrically, v
d
w
is represented in the pitch class torus by the path v + td, where t [0, 1]. For
6
example, voice leadings (i) and (ii) belong to the equivalence class (0, 4, 7)
(0,0,2)

(0, 4, 9), while (iii) belongs to the equivalence class (0, 4, 7)

(0,0,10)
(0, 4, 9).
The former voice leading takes the most ecient route on the pitch class torus
between the two points. In general, v
d
w and x
f
y are equivalent if and only
if d and f are identical and v and x represent the same point in the torus.
9
2.2 Distance and continuity
Continuity is a desirable property of transformations acting on a CQT space
for several reasons. Music theorists sometimes measure voice leadings using
distance functions reecting the aggregate amount of vocal changethat is, they
measure the line segment that represents the voice leading.
10
In order to help
the listener parse contrapuntal music into separate melodic streams, composers
often choose voice leadings that minimize the amount of vocal movement. In
CQT quotient spaces, the distance between two equivalence classes of pitch
vectors is the size of the smallest voice leading between their elements (that
is, CQT use the quotient metric). Points that are close in a CQT space are
essentially retunings of the same object; continuity ensures that transformations
act in similar ways on them. Moreover, continuous transformations do not map
paths representing voice leadings between nearby chords too far from each other.
Since traditional musical set theory considers a discrete musical universe, the
issue of continuity has not arisen outside of geometrical music theory.
3 Linear contextual transformations
There are three reasons to consider linear maps when modeling contextual trans-
formations: (1) since linear transformations preserve lines and line segments
represent voice leadings, they map one voice leading to another, (2) transfor-
mations employed by composers and commonly appearing in the music theory
literaturetransposition, inversion, retrograde (reversal of the order of pitches
in a vector), and permutationare linear transformations on ordered pitch or
pitch class space, and (3) many music theorists have already developed linear
transformation models for particular contextual transformations (see Lewins
treatment of serial transformations [12, sections 8.2.1, 8.3.1, 8.3.2, 9.6.4] and
Fiores representation of neo-Riemannian transformations [6]), though no gen-
eral theory has been proposed.
There are limitations to this approach, however. Linear transformations are
incapable of representing transformations such as set complementation that as-
9
Voice leadings in other CQT quotient spaces are more complicated. If f and g are OPTI
transformations (for example, permutations), the progression vw, where v, w R
n
, is
said to be uniformly equivalent to f(v)f(w) and individually equivalent to f(v)g(w).
Individually equivalent voice leadings correspond to paths connecting the same two points in
some CQT space. The paths may not coincideonly their endpoints must coincide.
10
If we make certain reasonable assumptions, the geometry imposed on R
n
by such distance
functions cannot deviate too far from Euclidean geometrysee [8] for details.
7
sume an equally tempered pitch universe. The fact that linear transformations
only apply to ordered multisets creates a more serious problem. In general,
points in CQT spaces represent equivalence classes of pitch vectors. For exam-
ple, chords are unordered multisets of pitch classes (points in R
n
/OP). There-
fore, linear transformations may be dened on elements of equivalence classes
modulo permutation but do not necessarily act in the same way on the entire
class. For this reason, we focus on linear transformations on R
n
or R
n
/O = T
n
.
This means that inversion, for example, is narrowly dened as the map v v,
rather than a more general map on chords.
3.1 The linear contextual group C
Contextual transformations are typically described as transposition- and inver-
sion-independent actionsthat is, group actions that commute with transpo-
sition and inversion. Moreover, contextual transformations should be, in some
sense, determined by the objects they act on. In general, contextual transfor-
mations that are dened in the music theory literature are functions determined
by the intervals (distances between elements) in a pitch vector, pitch class vec-
tor, or chord. They are not, in general, isometries of pitch or pitch class space.
We also assume that contextual transformations are invertible, meaning that
no nonzero subspace of R
n
or T
n
is mapped to zero. Moreover, we assume
that contextual transformations are well dened on ordered pitch class space
T
n
, meaning that if two pitch vectors are equivalent modulo octave shift, their
images under a contextual transformation are also equivalent modulo octave
shift.
Denition 1 The linear contextual group C is the group of linear transforma-
tions on n-dimensional ordered pitch space R
n
that are invertible, commute with
transposition and inversion, and are well dened on the pitch class torus T
n
.
We use matrix representations to investigate the structure of C and several of
its subgroups that arise in the music theory literaturein particular, subgroups
consisting of contextual transformations that preserve chord type or set class.
The ane linear transformation z Mz + b, where M is invertible, com-
mutes with transposition if and only if, for z R
n
and c a real number,
M(z + c1) +b = Mz +b + c1. (1)
Therefore, the commutativity condition requires that 1 is an eigenvector of
M with eigenvalue 1. We can think of an ane linear transformation that
commutes with transposition as the composition of a linear transformation z
Mz and a voice leading Mzv +b. The requirement that a transformation on
pitch space commute with inversion (multiplication by 1) means that b = 0 in
equation (1), and so linear contextual transformations are of the form z Mz.
Multiplication by M is well dened on the pitch class torus if
M(z + (0, . . . , 0, 12, 0, . . . , 0)) Mz (mod 12),
8
which implies that the entries of M are integers and hence the determinant
of M is 1. In other words, C is a subset of the group of invertible linear
transformations with integer entriesthat is, the general linear group GL(n, Z).
This gives us a complete characterization of the linear contextual group:
C = M GL(n, Z) : M1 = 1.
Note that linear contextual transformations constitute a discrete group because
they are required to be dened on the torus. Without this restriction, invertible
linear transformations that commute with transposition and inversion constitute
the (continuous) group of invertible matrices with real entries that x 1.
3.2 Matrix representation and multiplication of linear con-
textual transformations
We now explore compositions of linear contextual transformations with a view
towards understanding them from a geometric and group theoretic perspective.
It is convenient to choose a basis for R
n
so that 1 is a basis vector. Using the
basis
1, (1, 1, 0, . . . , 0), (1, 0, 1, 0, . . . , 0), . . . , (1, 0, . . . , 0, 1),
every pitch vector (z
0
, z
1
, . . . , z
n1
) can be written in the form (x
0
[ x
1
, x
2
, . . . , x
n1
)
where x
0
= z
0
, x
1
= z
1
z
0
, x
2
= z
2
z
0
, and so on. (Think of x
0
as the rst
note in the vector, and the other x
i
s as intervals measured from the rst note;
a C major triad in root position is represented (0 [ 4, 7), a D major triad in root
position is represented (2 [ 4, 7), and a C major triad in rst inversion is repre-
sented (4 [ 3, 8).) With this basis, every element of C can be written uniquely
in the form

1 a
1
a
n1
0
.
.
. A
0

= a, A) (2)
where (a
1
, . . . , a
n1
) = a is in Z
n1
and A is in GL(n 1, Z). Representing
matrices of the form (2) compactly as pairs a, A), we see that the action of C
on pitch space is given by
a, A)(x
0
[ x) = (x
0
+a x[ Ax).
Multiplication in C follows the rule
a, A) b, B) = aB +b, AB). (3)
Let id denote the (n 1) (n 1) identity matrix and 0 the 1 (n 1) zero
vector. Then 0, id) is the identity in C. The inverse of the element a, A) is
aA
1
, A
1
).
9
3.3 The structure of the linear contextual group
Theorem 1 The group C of linear contextual transformations is isomorphic to
the group of ane linear transformations on Z
n1
.
Proof. Let x be a vector in Z
n1
and let C be a matrix in GL(n 1, Z);
any pair (c, C) represents an ane linear transformation (c, C)x = c + Cx.
Composition of ane linear transformations denes a product on the set of
pairs (c, C):
(d, D) (c, C)(x) = d + Dc + DCx = (d + Dc, DC)x. (4)
Comparison of the equation (3) with the product dened in (4), (d, D)(c, C) =
(d + Dc, DC) proves the isomorphism. (Note that composition of ane linear
transformations acts like right multiplication in C.) 2
Note that (0 [ 1, 0, . . . , 0), (0 [ 0, 1, . . . , 0), etc. form a basis for R
n
/T R
n1
,
while (1 [ 0, . . . , 0) is the basis for a one-dimensional subspace. Geometrically,
we have decomposed R
n
into the product R R
n
/T. Any linear contextual
transformation of the form 0, A) denes a linear transformation on the space
of pitch vectors modulo transposition R
n
/T. This space represents sequences of
pitches sharing the same intervalsdierent transpositions of the same harmony
or melody, for example. Any point in the orbit of the pitch vector (0 [ x) lies in
the lattice of integer linear combinations of the projections of x onto the basis
of R
n
/T.
Lemma 1 The set CT = a, id) : a Z
n1
is a normal subgroup of C that
is isomorphic to Z
n1
.
Proof. Since a, id) b, id) = a+b, id), CT is a group isomorphic to Z
n1
.
Moreover, if g, G) is any element of C, then
g, G)
1
a, id) g, G) = gG
1
, G
1
) aG, +g, G) = aG, id) CT.
2
Therefore, the following sequence is exact:
0 Z
n1
C GL(n 1, Z) 0.
We see that C is isomorphic to the semidirect product Z
n1

GL(n 1, Z),
where the isomorphism : GL(n 1, Z) Aut(Z
n1
) is given by right multi-
plication by the elements of GL(n 1, Z).
3.4 Contextual transposition, inversion, and retrograde
This section is summarized in table 1.
10
3.4.1 Contextual transposition
An element of CT transposes a pitch vector (x
0
[ x) and any of its inversions
(y
0
[ x) in opposite directions by equal amounts (that is, a, id)(x
0
[ x) = (x
0
+
a x[ x) and a, id)(y
0
[ x) = (y
0
a x[ x)). Such transformations are called
contextual transpositions, or schritts [5]. The magnitude of the transposition is
[a x[. For example, the transformation
(0, 1), id) =

1 0 1
0 1 0
0 0 1

sends the trichord (x

0
[ t, f) to (x
0
+ f [ t, f). In other words, it transposes
the trichord by its second interval. We can write any major triad in the form
(x
0
[ 4, 7) and any minor triad in the form (x
0
[ 3, 7); these forms will be called
root position, as the rst coordinate of the vector identies the letter name
(root) of the triad, the middle coordinate identies the size of its third, and
the last coordinate identies the size of its fth. Thus, the transformation
transposes any triad in root position up by its fth. Note, however, that the
inversion of a root position triad is not in root position. For example, the
inversion of a root-position minor triad (x
0
[ 3, 7) is (x
0
[ 4, 7).
3.4.2 Contextual inversion
The group of linear contextual transpositions and inversions CTI consists of con-
textual transpositions and contextual inversionstransformations of the form
(x
0
[ x) (y
0
[ x). The element 0, id) generates a group of order two; it
inverts the intervals in a pitch vector while xing its rst note. We form the
exact sequence
0 Z
n1
CTI Z
2
0,
where CTI = a, id) C is the semidirect product Z
n1

Inv
Z
2
, where
Inv : Z
2
Aut(Z
n1
) maps Z
2
to id. The group CTI of contextual inversions
and transpositions is not normal in C. Elements of CTI act on points (x
0
[ x) :=
(x
0
[ x
1
, . . . , x
n1
) in ordered pitch space by
a, id)(x
0
[ x) = (x
0
+a x[ x).
In other words, the amount by which the vector or its inversion is transposed is
a sum of integer multiples of its intervals.
3.4.3 Contextual retrograde
The retrograde operation, in the standard basis for R
n
, is given by coordinate
reversal: (z
0
, z
1
, . . . , z
n1
) (z
n1
, z
n2
, . . . , z
0
). After the change of basis, it
is given by
(x
0
[ x
1
, . . . , x
n1
) (x
0
+x
n1
[ x
n2
x
n1
, x
n3
x
n1
, . . . , x
1
x
n1
, x
n1
).
11
Let R be the (n 1) (n 1) matrix
R =

0 0 1 1
.
.
. .
.
.
1 0 1
0 .
.
.
0 0 1
1 .
.
.
.
.
. .
.
.
.
.
.
0 0 0 1

Then retrograde is represented by (0, . . . , 0, 1), R); any transformation a, R)

can be thought of as a contextual retrograde. Since R
2
= id, the composi-
tion of two contextual retrogrades is a contextual transposition. Retrogrades
generate a group of contextual transpositions and retrogrades CTR that is a
semidirect product Z
n1

Ret
Z
2
, where the Z
2
action on Z
n1
is left multipli-
cation by R. Although both CTI and CTR are semidirect products of Z
n1
and Z
2
, they are not isomorphic:
Lemma 2 The contextual transposition/inversion group CTI is not isomorphic
to the contextual transposition/retrograde group CTR.
Proof. Every element of CTI is either a transposition a, id) or an involution
a, id) (since a, id)
2
= a a, id) = 0, id)). However, elements a, R) are
involutions only if a = aR. 2
3.5 The extended linear contextual group C
Linear contextual transformations are not the only invertible linear transforma-
tions that are well dened on pitch class space and commute with transposition
and inversion. If z is a pitch class vector, then M(z) + b (Mz + b)
(mod 12) if and only if b (6Z)
n
. We enlarge the linear contextual group to
include transpositions that commute with inversion in pitch class space:
Denition 2 The extended linear contextual group is the group of linear trans-
formations of n-dimensional ordered pitch class space T
n
that are invertible and
commute with transposition and inversion.
Therefore,
C = (b, M) : M GL(n, Z), M1 = 1, and b (6Z)
n

where the action of (M, b) is dened by (M, b)(z) = Mz +b. It is not dicult
to show that (b, id) : b (6Z)
n
is normal in C; therefore C (6Z)
n
C.
However, as contextual is generally taken to exclude transpositions, this group
is not one appearing in music theory literature.
12
Group Description Structure Comments
C Invertible linear transformations on ordered pitch space that
commute with transposition and inversion and are well dened on
T
n
Z
n1
GL(n 1, Z);
GL(n 1, Z) acts by left
multiplication.
Isomorphic to the group of ane
linear transformations of Z
n1
.
T Linear contextual transformations that x the rst pitch in a
vector; invertible linear transformations of the ordered
transposition class torus R
n
/T R
n1
GL(n 1, Z) Used in tuning theory.
CT Transpositions by linear combinations of the intervals in a pitch
class vector
Z
n1
Normal in C
CTI Generated by contextual transposition and inversion about the
rst pitch class in a vector
Z
n1

Inv
Z
2
;
Z
2
= id, id
Contains a representation of the
PLR group as a subgroup
(n = 3).
CTR Generated by contextual transposition and retrograde Z
n1

Ret
Z
2
;
Z
2
= id, R
Not isomorphic to CTI.
Contains generators of Hooks
UTT group (n = 3).
CTIR Generated by contextual inversions, retrogrades, and retrograde
inversions
Z
n1
(Z
2
Z
2
);
(Z
2
Z
2
) =
id, id, R, R
The serial transformation
group.
Table 1: Summary of linear contextual transformations.
1
3
4 Orbits of linear contextual group actions
As previously discussed, the orbits of pitch vectors under the actions of the
OPTI groups have particular musical signicance (they represent chords, chord
types, etc.). The orbit of a pitch or pitch class vector under the action of the
linear contextual group can be thought of as a set of pitch vectors that are
in tune with each other in the following sense. Suppose we choose any two
members of this orbit, v and w, and construct any voice leading of the form
v
d
w. Then the coordinates of d are linear combinations of the intervals in v
(modulo octave equivalence in the pitch class case). For example, consider the
orbit of the just tempered A major triad v = (9 [ , ), where = 12 log
2
5/4
and = 12 log
2
3/2. The orbit is
(9 + a + b [ c + d, e + f : a, b, c, d, e, f Z and cf de = 1.
Members of this orbit include not only obvious examples like the A minor triad
(9 [ , ) but also vectors like (9 [ , +4) (in pitch class space, the intervals
of this vector are a perfect fth and a syntonic comma). The restriction cfde =
1 ensures that each vector is the image of v under the action of an invertible
linear transformationessentially, that each vector in the orbit denes the same
tuning system. (Lifting the restriction produces a complete set of pitch (class)
vectors w such that the coordinates of wv are integer linear combinations of
the intervals in v.) If the intervals in a pitch or pitch class vector (x
0
[ x) have the
property that the equation a x = 0 has no nonzero solutions in Z
n1
(or Z
n1
12
for a pitch class vector), then we say that the intervals are incommensurate. If
the intervals of a pitch class vector (x
0
[ x) are incommensurate, then its orbit
under the action of C is dense in the subspace x
0
0 of R
n
/O. Musically, this
means that every tuning system with incommensurate intervals is capable of
approximating any chord type to within any given degree of accuracy.
Lewin [12] dened a generalized interval system to be a group of intervals
acting simply transitively on a set of musical objects.
11
Ordered pitch class space
is a topological group that acts simply transitively on itself by vector addition:
given two pitch class vectors x and y, there exists only one interval, namely
yx (mod 12), that takes x to y. In contrast, linear contextual transformations
do not act transitively on ordered pitch space or the ordered pitch class torus
for one, every element of C xes the origin. However, C does act transitively on
the orbits of some pitch or pitch class vectors. Let v = (x
0
[ x). Since the matrix
form of a linear contextual transposition has integer entries, A, a)(x
0
[ x) =
(x
0
[ x) has nontrivial solutions if and only if the intervals of v (the coordinates
of x) are not incommensurate. Therefore, given a pitch vector whose intervals
are incommensurate, we can form a generalized interval system whose group of
11
A group G acts simply transitively on a set S if for every pair (s, t) of elements of S
there exists exactly one element g of G such that gs = t. Roughly, we can think of a simply
transitive action as an unambiguous: there is exactly one action on s that produces t. The
stabilizer of an element s S is the subgroup of elements g G that x s (i.e., {g : gs = s}).
The group G acts simply transitively on the orbit of s S if and only if the stabilizer of s is
trivial.
14
intervals (in the Lewinnian sense) is C. For example, the orbit of the just
tempered A major triad (9 [ , ) described above forms a generalized interval
system with C as its group. We can also construct a generalized interval system
using one of the subgroups of C. For example, the orbit of the just tempered
triad under the action of the group of contextual transpositions and inversions
(CTI) is the set
(9 + a + b [ , : a, b Z and = 1.
This set is a generalized interval system with CTI as its group of intervals. It
contains major triads ( = 1) and minor triads ( = 1).
Musical set theory often considers only equally tempered setsthat is, those
whose intervals are rational. Every pitch or pitch class vector with equally
tempered intervals is xed by some nontrivial subgroup of C. For example, all
12-tet vectors
12
are xed by 12a, id). Each individual 12-tet vector is xed by
additional transformations. For example, in pitch class space, (x
0
[ 4, 7) is xed
by any transformation k(1, 8) [ id) where k Z
12
. The quotient of CTI by the
stabilizer subgroup of (x
0
[ 4, 7) is isomorphic to Z
12
Z
2
(the dihedral group
D
24
, otherwise known as the group of neo-Riemannian transformations in equal
temperament).
4.1 Distance
Even on the pitch class circle, there are at least two competing notions of size at
work. Geometrical music theory normally measures Euclidean distance on the
continuous pitch class circle, while tuning theorists dene a distance based (for
example) on Pythagorean tuning: the distance between two pitch classes is the
number of pure fths (a 3/2 frequency ratio, or 12 log
2
3/2 units of pitch) from
one to the other, with the distance between two pitch classes that are unrelated
by pure fths being undened. Therefore, the distance between pitch classes
separated by four Pythagorean commas
13
that is, separated by an interval
approximately equal to a semitoneis 48, while the size of an equally tempered
semitone is undened.
When we take voice leadings into account, the actions of contextual linear
transformations are anything but well behaved. In general, they are not isome-
tries of R
n
or T
n
. In fact, for any pitch vector z, there exists a pitch vector w
arbitrarily close to z that is mapped arbitrarily far from z by repeated applica-
tion of some linear contextual transformation a, A) to both vectors. Therefore,
although voice leadings (line segments) are mapped to other voice leadings,
their Euclidean length is not preserved. This is rather discouraging, as one of
the most attractive features of CQT spaces is that they possess a natural notion
of distance based on minimal voice leadings.
12
A 12-tet vector is one whose intervals may be embedded into 12-tet: (x
0
| x) is a 12-tet
vector if x 0 (mod 12). The value of x
0
is immaterial.
13
A Pythagorean comma is the distance between pitch classes separated by twelve pure
fths (12 log
2
3
12
/2
19
). Since a pure fth is nearly 7/12 of an octave, this distance is small
approximately 0.2346 semitonesbut noticable even to an untrained ear.
15
5 Musical examples
We consider specic musical examples in which the linear transformation model
is protable.
5.1 Sequences of fths
Let the contextual transposition F equal (1), id) acting on T
2
. Successive
application of F to a two-element pitch class vector (that is, an interval) (x
0
[ g)
gives the sequence of fths (x
0
+ g [ g), (x
0
+ 2g [ g), (x
0
+ 3g [ g), . . .. In the
pitch class circle, the orbit of the equally tempered major third (g = 4) under
the action of F has three points and the orbit of the equally tempered perfect
fth (g = 7) has twelve points. If g is an irrational number, such as the pure
fth 12 log
2
3/2, the map g ng (mod 12) is a chaotic dynamical system and
the orbit of (x
0
[ g) under the repeated action of F is dense in the subspace
spanned by (k [ g) for k in R/12Z.
5.2 Linear contextual transpositions and tuning theory
Tuning theorists ignore transposition level and study the eect of linear trans-
formations on the intervals in a pitch class vector. In other words, they consider
pitch class segments (x
0
[ x) and (y
0
[ y) to dene the same tuning if there exists
an invertible linear transformation A GL(n1, Z) such that y = Ax [13]. The
tuning group T is the quotient of C by the set of contextual transpositions
that is, the general linear group GL(n 1, Z). It is isomorphic to the mapping
class group of (n 1)-dimensional torus, or ordered ordered interval space
R
n
/OT T
n1
.
5.3 Neo-Riemannian transformations.
The CTI group acting on R
3
or T
3
is generated by the linear transformations
W
12
= (1, 0), id), W
13
= (0, 1), id), and W
23
= (1, 1), id). Each W
ij
is a contextual inversion that exchanges the ith and jth elements of the pitch
class vector. It has the presentation W
12
, W
13
, W
23
: W
2
12
= W
2
13
= W
2
23
=
(W
12
W
13
W
23
)
2
= 1). When restricted to 12-tet triads, W
12
, W
13
, and W
23
act as R, P, and L, respectively.
14
The tonnetz (gure 3) is the standard geo-
metrical representation of the action of the PLR group on triads. (Ignore the
numerals after the chord namesthey are needed for a construction discussed
in section 5.4.) If the node labels are erased, the resulting honeycomb is the
14
Fiore used this representation in an unpublished article [6]. In their article on generalized
contextual groups [5], Fiore and Satyendra dene contextual transformations on pitch class
vectors dierently. A schritt Q
k
for k Z
12
is a contextual transposition such that Q
k
y =
y + k1 if y is a transposition of some given vector v Z
n
12
and Q
k
y = y k1 if y is a
transposition of a retrograde-inverted form of v. A wechsel, K, is similar to W
12
(as dened
in section 5.3); it sends y to its retrograde inverted form sharing the same rst two pitch
classes. This construction cannot be extended to continuous space, because Q
k
is undened
on vectors that are invariant under retrograde inversion.
16
Cayley graph of the PLR group (and CTI group), with parallel line segments
representing the same transformation.
15
In gure 3, P is vertical, R is north-
west/southeast, and L is northeast/southwest. The CTI group does not act
simply transitively on the orbit of any pitch class vector that can be embedded
in equal temperament. Even in 12-tet, the size of an orbit depends on the vector
chosen: for example, the orbit of the augmented triad (x[ 4, 8) can have no more
than six elements, since the augmented triad can be embedded in 3-tone equal
temperament. In contrast, orbits of pitch class vectors with incommensurate
intervals are innite; in this case, CTI acts simply transitively.
5.4 Parsimonious transformations
Cohn [3] recognized the importance of small voice leadings in composers im-
plementations of neo-Riemannian transformations. He described neo-Riemannian-
type transformations acting on generic three-note set classes. In each case, two
voices are xed, while the remaining voice moves as needed to invert the chord.
We can express this phenomenon with matrices. In ordered pitch class space,
the x transformations F
12
, F
13
, and F
23
, where F
ij
xes the ith and jth
voices, have the following matrix forms:
F
23
(z) =

1 1 1
0 0 1
0 1 0

z F
13
(z) =

1 0 0
0 1 1
0 0 1

z
F
12
(z) =

1 0 0
0 1 0
0 1 1

z.
For example, F
23
(x[ 4, 7) = (x1 [ 5, 8) (that is, F
23
maps a major triad in root
position to a minor triad in second inversion), while F
23
(x[ 5, 9) = (x + 2 [ 3, 7)
(that is, F
23
maps a major triad in second inversion to a minor triad in root
position). The maximally even set (x[ 4, 8) is xed by all F
ij
. Voice leadings
zF
ij
(z) are parsimonious when z is close to an augmented triad.
Like the tonnetz, the Cayley graph of the group generated by the F
ij
s is
a honeycomb. However, its edges are labeled dierently (see gure 3). The
x transformations are indicated by the style of line segments connecting
nodes; each line segment represents both a contextual transformation and a
voice leading, which we interpret as z
d
F
ij
(z), where each d
i
is the smallest-
magnitude member of its equivalence class.
Keeping track of voice leadings reveals signicant musical dierences be-
tween the PR, LP, and LR chains described by Cohn [3]. All of these produce
loopssequences of pitch class vectors cycling through the same orbitwhen
applied in equal temperament. However, only the LP cycle is a sequence of min-
imal voice leadings that can be realized by simultaneous melodies in which each
15
Cayley graphs are not unique, since a group does not have a unique set of generators.
Here, the generators are assumed to be P, L, and R.
17
F13
RL
static
PR
descending
PL
static
ascending
F12
R
L
transformations
Fix
transformations
Neo-Riemannian
P
F23
C0
a1
F2
A1
b0
g

1
B0
D2
d2
B

0
d0
D0
f

2
c

0
f

1
F

1
d

2
B

1
b

1
D

0
e

2
b

0
G

1
C

0
a

1
E

2
F

2
e

2
G

1
a

1
d

0
b

0
D

2
C

0
F

2
E2
f2
A

1
c0 g1
G1
e2
Figure 3: Transformational network depicting the x transformation group
acting on consonant triads. Recall that capital letters indicate major triads and
lowercase letters indicate minor triads. The numerals 0, 1, and 2 indicate root
position, rst inversion, and second inversion, respectively.
voice returns to its starting pitch and no voice leaps by more than a whole tone.
The other two chains involve parsimonious voice leadings producing melodies
that ascend or descend in pitch as indicated. Cohns example 1 [3] (Brahms,
Concerto for Violin and Cello, Op. 102, First Movement, mm. 270-78) shows
Brahms using the LP loop to eect: the violin parts traverse a static cycle.
Although the viola leaps by an octave in the fourth progression, this is clearly
not forced by voice-leading considerations. (The cello part is driven by fun-
damental bass rules rather than voice leadings.) All together, the string parts
create a bed above which the two solo parts soar. In contrast, the overall
eect of a parsimonious circle of fourths progression is an ascending line, as in
measures 14-19 of di Lassos Carmina Chromatico from Prophetiae Sibyllarum
(gure 5).
Parsimonious transformations also reveal some interesting relationships in
Schuberts A major reharmonization of the A minor theme by Anselm H utten-
brenner in variation XIII (gure 6). The similarity between the progression G
majorC major A minor in the theme and G

major C

minor A major
in variation XIII is driven by voice leadingsby x transformationsrather
than neo-Riemannian duality.
This example also demonstrates the problems we encounter when measuring
distances. One normally denes distance on a Cayley graph using the word
length metric, where the distance between vertices is the smallest number of
edges connecting them. This distance is not generally the same as distance in
18
&
?
Violin 1 & 2
Viola & Cello

A
b
2

b
b

#
g
#
2

#
#

#
E0

n
n

n
e0

C1

c1

b
A
b
2

b
b
Figure 4: Brahms, Concerto for Violin and Cello, Op. 102, First Movement,
mm. 270-78. With the exception of the octave leap in the viola, the upper three
voices are relatively static.
a CQT space. For example, although c0 and a1 are equidistant from C0 on the
tonnetz, c0 represents only one semitones worth of movement from C0, while a1
is two semitones away. When applied to triads with incommensurate intervals,
the distance functions dier widelyroughly, distance on the tonnetz is related
to the number of commas by which two triads dier, while geometric distance
is the size of the smallest voice leading from one chord to another.
5.5 Uniform triadic transformations.
Tonal theory has a particular interest in groups of triadic transformations,
including the neo-Riemannian transformations. Hook [9] proposed the 288-
element group of uniform triadic transformations (UTTs), which is isomorphic
to (Z
12
)
2
Z
2
. The UTT group acts on the set of 12-tet minor and major triads
and contains two types of elements: those that preserve mode (i.e. minor or
major) and those that reverse mode. Any pair (t
+
, t

) Z
2
12
and +, de-
nes a unique triadic transformation (, t
+
, t

): indicates whether the trans-

formation is mode-preserving (+) or mode-reversing (), and t
+
and t

indi-
cates the transposition level of major and minor triads, respectively. So, for
example, the neo-Riemannian R equals (, 3, 3) and the dominant transfor-
mation equals (+, 5, 5).
The structural similarity between the UTT group and the linear contextual
group in 12-tet suggests that we can construct a matrix representation of Hooks
group using linear contextual transformations. Let A =

1 1
0 1

. A solution
to the system of linear equations
4a + 7b t
+
(mod 12)
3a + 7b t

(mod 12)
is a t
+
t

(mod 12); since 7 is relatively prime to 12, the equation 7b

t
+
4a (mod 12) has a unique solution in Z
12
. The UTT (, t
+
, t

) may be
then represented by (a, b), id) if = + and (a, b), A) if = , where (a, b)
19
&
?
2
4
2
4
SA
TB
# w
w #
w #
F
#
1 B0
w #
w
#
w
W #
w
#

E2
w
w #
. w

w #

A0 D2
w
w
w #

n w

w n
G1 C0 G1
W n
.
n

n


w

w
C0 F2

w
w

b
.

#

.

n
D0 g2 E
b
0 B
b
1

. b
b

.
b

Figure 5: The soprano, alto, and tenor parts participate in an ascending circle of fourths progression in di Lassos Carmina
Chromatico, mm. 14-19.
&
&
Theme
mm. 1-8
Var. XIII
mm. 79-86

a0

#
A0

#
E2

#
E2

a0

#
A0

d2

#
#
#
???

n
G0

#
#
#
G
#
0

G1

#
#
#
G
#
1

C0

#
#
c
#
0

a1

#
A1

a1

#
A0

#
E2

#
E1

a1

#
A0
R
F
13
F
23
RL
L
F
12
F
12
F
23
F
13
F
23
PLR
Figure 6: Schuberts reharmonization from Variations on a theme by Anselm H uttenbrenner. In variation V, a D major
chord substitutes for the diminished chord marked ???.
2
0
(Z
12
)
2
. For example, the UTT D = (+, 5, 5) is represented by (0, 11), id) (any
transformation (0, b), id) transposes major and minor triads by equal amounts).
In this system, the neo-Riemannian transformations are represented by P

, L

,
and R

, where P

= (0, 0), A), L

= (8, 8), A), and R

= (6, 3), A).

5.6 Serial transformations
Contextual inversion and retrograde are favorites of serial composers. David
Lewins GMIT presents numerous examples of contextual transpositions, inver-
sions and retrogrades acting on series (pitch class vectors). Examples include
RICH (GMIT, section 8.2.1), TCH = RICH
2
(section 8.2.1), MUCH (8.2.5),
BIND (9.6.4), TLAST (8.3.1), TFIRST (8.3.1), FLIPEND (8.3.2), and FLIP-
START (8.3.2) [12]. Of these, only one, MUCH, is non-linear (MUCH takes a
pitch class vector s to its retrograde-inverted form whose beginning overlaps the
ending of s as much as possible). With the exception of TCH, each transfor-
mation preserves at least one common tone. For example, RICH, which takes a
series s to its retrograde-inverted form beginning with the last two elements of s,
is a linear contextual retrograde inversion represented by (0, . . . , 0, 1, 0), R).
FLIPEND and FLIPSTART are rare examples of linear contextual transforma-
tions that do not preserve set class. FLIPSTART takes a pitch class vector
(s
1
, s
2
, s
3
) in R
3
/O to (a, s
1
, s
3
), where a is the inversion of s
2
about s
1
, and
FLIPEND is constructed in a similar manner.
Conclusion
This paper demonstrates that the actions of contextual transformations on con-
tinuous ordered pitch or pitch class space can be represented by matrix multipli-
cation. In doing so, it connects tuning theory with discrete musical set theory
and gives new insight into foundational constructions such as serial transfor-
mations and the neo-Riemannian transformations. We have only scratched the
surface of what this model can reveal about the structure of contextual transfor-
mation groups. However, we must keep in mind that the constructions developed
in this paper are not the only model for contextual transformations; requiring
transformations to be dened on ordered sets is problematic in some musical
contexts.
Acknowledgements.
I thank John Hall, Erik Van Erp, and Dmitri Tymoczko for their helpful com-
ments.
References
[1] Clifton Callender, Ian Quinn, and Dmitri Tymoczko. Generalized voice-
leading spaces. Science, 320:346348, 2008.
21
[2] John Clough. A rudimentary geometric model for contextual transposition
and inversion. J. Mus. Theory, 42(2):297306, Autumn, 1998. Special issue
on Neo-Riemannian Theory.
[3] Richard Cohn. Neo-Riemannian operations, parsimonious trichords, and
their tonnetz representations. J. Mus. Theory, 41(1):166, 1997.
[4] Leonhard Euler. De harmoniae veris principiis per speculum musicum
repraesentatis. In Novi commentarii academiae scientiarum Petropolitanae,
pages 330353, St. Petersburg, 1774.
[5] Thomas Fiore and Ramon Satyendra. Generalized contextual groups. Mus.
Theory Online, 11(3), 2005.
[6] Thomas M. Fiore. Music and mathematics, 2008. URL:
http://www.math.uchicago.edu/ ore/1/musictotal.pdf. Retrieved Jan-
uary 3, 2008.
[7] Edward Gollin. Some aspects of three-dimensional tonnetze. J. Mus.
Theory, 42(2):195206, Autumn, 1998. Special issue on Neo-Riemannian
Theory.
[8] Rachel W. Hall and Dmitri Tymoczko. Poverty and polyphony: A connec-
tion between economics and music. In Reza Sarhangi and Carlo Sequin,
editors, Bridges: Mathematical Connections in Art, Music, and Science,
Donostia, Spain, 2007.
[9] Julian Hook. Uniform triadic transformations. PhD thesis, Indiana Uni-
versity, Bloomington, IN, 2002.
[10] Jonathan Kochavi. Some structural features of contextually-dened inver-
sion operators. J. Mus. Theory, 42(2):307320, Autumn, 1998. Special
issue on Neo-Riemannian Theory.
[11] David Lewin. A formal theory of generalized tonal functions. Journal of
Music Theory, 26(1):2360, 1982.
[12] David Lewin. Generalized Musical Intervals and Transformations. Yale
University Press, New Haven, CT, 1987.
[13] Andrew Milne, William A. Sethares, and James Plamondon. X system,
2006. URL: www.thummer.com/ThumTone/X System.pdf. Retrieved
April 24, 2008.
[14] A. v. Oettingen. Harmoniesystem in Dualer Entwicklung. W. Glaser,
Leipzig, 1866.
[15] Hugo Riemann. Skizze einer Neuen Methode der Harmonielehre. Breitkopf
& Hartel, Leipzig, 1880.
[16] Dmitri Tymoczko. The geometry of musical chords. Science, 313(5783):72
74, 2006.
22