The Continuous Hexachordal Theorem

Brad Ballinger1, Nadia Benbernou2 , Francisco Gomez3 , Joseph O’Rourke4 ,

and Godfried Toussaint5
1
Davis School for Independent Study, Davis, USA
2
Department of Mathematics and Computer Science and Artificial Intelligence
Laboratory (CSAIL), Massachusetts Institute of Technology, Cambridge, USA
3
Department of Applied Mathematics, Universidad Politecnica de Madrid,
Madrid, Spain
4
Department of Computer Science, Smith College, Northampton, USA
5
Centre for Interdisciplinary Research in Music Media and Technology,
The Schulich School of Music, McGill University, Montreal, Canada

Abstract. The Hexachordal Theorem may be interpreted in terms of

scales, or rhythms, or as abstract mathematics. In terms of scales it
claims that the complement of a chord that uses half the pitches of a
scale is homometric to—i.e., has the same interval structure as—the orig-
inal chord. In terms of onsets it claims that the complement of a rhythm
with the same number of beats as rests is homometric to the original
rhythm. We generalize the theorem in two directions: from points on a
discrete circle (the mathematical model encompassing both scales and
rhythms) to a continuous domain, and simultaneously from the discrete
presence or absence of a pitch/onset to a continuous strength or weight of
that pitch/onset. Athough this is a significant generalization of the Hex-
achordal Theorem, having all discrete versions as corollaries, our proof
is arguably simpler than some that have appeared in the literature.
We also establish the natural analog of what is sometimes known as
Patterson’s second theorem: if two equal-weight rhythms are homomet-
ric, so are their complements.

1 Introduction

1.1 Basic Definitions

We are concerned with cyclic musical rhythms consisting of k onsets (pulses,

beats) and n−k rests, represented by n evenly spaced points on a circle, with
arithmetic mod n, i.e., in the group Zn . This representation has been used as
early as the 13th century, as accounted by Wright [Wri78], but it has been used
recently again; see [Vuz85], [Tou05], among others. Alternately, the k onsets
(points) may be considered as k pitches making up a musical chord or scale
selected from a universe of n pitches [Tym06]. Such sets of points on a circle are
called cyclotomic sets in the crystallography literature [Pat44], [Bue78]. We will
emphasize the rhythms model in this paper, but all results hold equally in the
pitch model or the crystallography model.

E. Chew, A. Childs, and C.-H. Chuan (Eds.): MCM 2009, CCIS 38, pp. 11–21, 2009.

c Springer-Verlag Berlin Heidelberg 2009
12 B. Ballinger et al.

Every pair of the points on the circle determines an inter-onset duration in-
terval (the geodesic between the pair of points around the circle) [Bue78]. The
histogram of this multiset of distances in the context of musical scales and chords
is called its interval content [Lew59]. Two rhythms which are congruent to each
other obviously have the same interval content. Here by congruence we mean
geometrical congruence, i.e., equivalence under rotation or reflection. However,
two rhythms with the same histograms need not be congruent. Two sets of points
with the same multiset of distances are said to be homometric, a term introduced
by Patterson in 1939 [Pat44], who first discovered them. In the music literature,
two pitch-class sets (or two rhythms) with the same intervalic content are termed
as having the Z-relation or isomeric relation [For77].
One of the fundamental theorems in this area is the so-called Hexachordal
Theorem, which states that complementary sets with k=n/2 (and n even) are
homometric. Two examples are shown in Figs. 1 and 2. In Fig. 1, the k=4 onsets
occur at (0, 1, 4, 7), and the complementary rhythm has onsets precisely where
the first rhythm has rests: (2, 3, 5, 6). The histogram of intervals is identical.
Fig. 2 shows two complementary (n, k)=(12, 6) rhythms, again with identical
histograms.
An important convention we follow is that the pair of onsets separated by the
diameter d = n/2 contributes two counts to the interval d in the histogram. This

1 1
7 1
2 2.0

6 4 2 1.0

3 3

1 2 3 4
5 3

7 1
2

6 4 2 1

1 3 3 1

1 2 3 4
2
5 3

Fig. 1. Example of the Hexachordal Theorem, (n, k)=(8, 4). Note that the distance
d=4 is counted twice.
 The Continuous Hexachordal Theorem 13

0
11 1 1

4
10 2
3
4
3
4
5
5 2
9 4 6 3
5

1 1
4
3
8 4
3 1 2 3 4 5 6
2

7 1 5
6

0
11 1

3 1 4
10 2
4
4
3
1 5
5

9 2 6 4 3 2

5 3
1 1
4
8 4
3 1 2 3 4 5 6

7 5
6

Fig. 2. Another example of the Hexachordal Theorem, (n, k)=(12, 6). Note that the
distance d=6 is counted twice.

convention simplifies the proofs but changes nothing substantively. This issue is
further addressed in Section 2.5.
The term “hexachordal” derives from Schönberg’s use of 6-note chords in a
12-tone chromatic scale, and the name “hexachordal” has been retained even
though the theorem holds for arbitrary even n.

1.2 History

The earliest proof of the Hexachordal Theorem in the music literature is, to our
knowledge, due to Lewin. In 1959 he published a paper [Lew59] on the intervalic
relations of two chords that contained an embryonic proof of the Hexachordal
Theorem; such a proof was refined in a subsequent paper [Lew60]. In 1974 Re-
gener [Reg74] found an elementary simple proof of this theorem based on the
combinatorics of pitch-class sets. Many other proofs have appeared since then,
often rivalling in conciseness. Short proofs can be found, for instance, in the
work of Mazzola [Maz03] or Jedrzejewski [Jed06]. Amiot [Ami07] gave an ele-
gant, short proof based on the discrete Fourier transform. Perhaps, one of the
simplest proofs, in the sense of using no structures such as groups or discrete
Fourier transforms, was discovered by Blau [Bla99]. His proof relies on a straight-
14 B. Ballinger et al.

forward analysis of the situation in which two complementary hexachords switch

two neighbouring elements.
The music theorists appear to be unaware that this theorem was known
to crystallographers about thirty years earlier [Pat44]. It seems to have been
proved by Patterson [Pat44] around 1940, but he did not publish a proof. In
the crystallography literature the theorem is called Patterson’s second theo-
rem [Bue76]. The first published proof in the crystallography literature is due
to Buerger [Bue76]; it is based on image algebra, and is non-intuitive. A much
simpler and elegant elementary proof was later found by Iglesias [Igl81]. An-
other simple proof, purely based on geometry, has been recently discovered by
Senechal [Sen08].
The Hexachordal Theorem has been generalized in various ways, for exam-
ple, considering rhythms of different cardinalities; see [Lew76], [Lew87], [Igl81],
[Mor90], [Sod95], [AG00] for several directions of generalization. We believe the
proof we present in Sec. 2.3 below is not only simple, but also establishes a
significant generalization from discrete rhythms to continuous rhythms.

1.3 Outline
We will first introduce weighted rhythms as a generalization of usual rhythms.
This generalization will consist of associating certain weights to the onsets and
rests of a rhythm. Next we will state and prove the Hexachordal Theorem in
terms of such weighted rhythms. We will then generalize the Hexachordal Theo-
rem to a continuous version of it , where rhythms will be considered as continuous
functions on the interval [0, 1]. From this version we will prove again the discrete
Hexachordal Theorem as a straightforward corollary of the continuous version.

2 The Continuous Hexachordal Theorem

2.1 Weighted Rhythms
In order to state our generalization of the Hexachordal Theorem, we introduce a
different viewpoint. Each onset i is assigned a weight of wi = 1, and each rest is
assigned a weight of 0. Thus, the rhythm in Fig. 1 (top) has a weight signature
n−1
(1, 1, 0, 0, 1, 0, 0, 1). The total weight of a rhythm R is W (R) = i=0 wi , the
number of onsets k in R. The complementary rhythm R is obtained by comple-
menting the weights with respect to 1: wi = 1 − wi . Let HR be the histogram
of intervals determined by rhythm R. This records, for each possible interval
distance d, the number of times it occurs in the rhythm. In Fig. 1, we have:

Height: 2 1 2 2
Distance d: 1 2 3 4

This may be viewed as a function of the interval distance d: HR (d) is the height
of the histogram at distance d. With this notation, the Hexachordal Theorem
may be stated as follows:
 The Continuous Hexachordal Theorem 15

Theorem 1. If R is a rhythm on n points, n even, and W (R) = n/2, then R

and R are homometric: for all distances d, HR (d) = HR (d).
Before proceeding to the continuous domain, we need Lemma 1 below, which
expresses the histogram function in terms of the weights. This lemma is known
in the music literature as the “common-tone theorem” [Joh03]. See [JK06] for a
proof in the context of group theory. For the sake of completeness, we include
our own proof.
n−1
Lemma 1. HR (d) = i=0 wi wi+d .
Proof. Point i is separated by a distance d from the point at i+d, where we
interpret addition mod n, i.e., in Zn . If both are onsets, then wi = wi+d = 1,
and wi wi+d = 1. If either point is a rest, then wi wi+d = 0. Thus, for each fixed
d, summing wi wi+d over all i counts 1 for each occurrence of d.
We now argue that each pair of points realizing a distance d contributes just
once to the sum. A pair (i, i + d) would contribute twice if i + 2d = i so that
(i+d, i) would be counted as well. Because d is a shortest path, we have d ≤ n/2.
Thus, i + 2d ≤ i + n, and this equals i (in Zn ) only when d = n/2 is the diameter.
Our convention is indeed to count a pair realizing the diameter twice.
Consider, for example, the n = 12 example in Fig. 2 (top). For d = n/2 = 6,
both w0 w6 and w6 w12 =w6 w0 contribute to HR (6) = 2. Indeed, the reason we
follow the convention of double-counting each realization of the diameter is that
it naturally fits this weight viewpoint. This point will be revisited in Section 2.5.

2.2 The Continuous Generalizations

We generalize in two directions. First, the circle of n discrete points is generalized
to a continuous circle of points. We take its circumference to be 1 without loss of
generality. Second, the discrete set of weights wi is generalized to a real-number
weight f (x) ∈ [0, 1] for x ∈ [0, 1]. Here x specifies a point on the circle, measured
by distance clockwise from the zero-position (conventionally at the 12 o’clock
position as in Figures 1 and 2), and f (x) the weight of that point. So now the
1
total weight W (R) = 0 f (x) dx. Note the maximum possible total weight of
any rhythm is achieved by the constant “rhythm” with weight f (x) = 1 for all
x, in which case W (R) = 1.
We define the complement of a rhythm analogously to the discrete case:
Definition 1. For each point x in rhythm R with weight f (x), the corresponding
point x in the complementary rhythm R has weight f (x) = 1 − f (x).
The histogram HR (d) is generalized to a function over the domain d ∈ [0, 12 ].
We need the continuous analog of Lemma 1. In fact, we take the analog of that
lemma as the definition of the histogram in the continuous domain:
1
Definition 2. HR (d) = 0 f (x)f (x + d) dx.
For example, if two points x and x + d each have weight 12 , they contribute 1
4 to
the height of HR at distance d.
16 B. Ballinger et al.

2.3 Continuous Hexachordal Theorem and Proof

The Continuous Hexachordal Theorem says that for any rhythm on the continu-
ous circle as described above, if the rhythm has weight 12 , then it is homometric
to its complement. More formally, it may be stated as:
Theorem 2. If R is a integrable rhythm on the continuous circle, and W (R) =
1
2 , then for all distances d, HR (d) = HR (d).

Proof. The proof fixes d and establishes that HR (d) = HR (d). From the his-
togram Definition 2 we have:
 1
HR (d) = f (x) f (x + d) dx.
0

From the complement Definition 1 this is:

 1
= [1 − f (x)][1 − f (x + d)] dx.
0

Multiplying out terms yields:

 1
= (1 − f (x) − f (x + d) + f (x)f (x + d)) dx.
0

Separating integrals gives:

 1  1  1  1
= 1 dx − f (x) dx − f (x + d) dx + f (x)f (x + d) dx
0 0 0 0

1
The first integral is just 1, and the second two1 are each 2 by the assumption of
the theorem that W (R) = 12 :
 1
1 1
=1− − + f (x)f (x + d) dx
2 2 0
 1
= f (x)f (x + d) dx
0
= HR (d)

The last step again follows from the Definition 2, and so we have established that
HR (d) = HR (d) for all d, i.e., the histograms are identical and R is homometric
to R.
The weight function f (x) need not be a continuous function in the technical
mathematical sense.2 We only need that it be integrable,3 i.e., a function for
which an appropriate “area under the function graph” may be defined.
1
Shifting x to x + d shifts the graph of f ( ) but does not change the area underneath it.
2
A function f is continuous if, for all c in the domain, limx→c f (x) = f (c).
3
For example, Lebesgue integrable suffices.
 The Continuous Hexachordal Theorem 17

f(x)
1

(a) x
0 1/8 1/4 3/8 1/2 5/8 3/4 7/8 1

H(d)
1/2

3/8

1/4

1/8

d
(b) 0 1/8 1/4 3/8 1/2

Fig. 3. (a) Weight step function f (x) corresponding to Fig. 1 (top), (n, k)=(8, 4).
(b) Corresponding histogram integral H(d).

We should note that the above proof can be directly discretized to yield a
parallel proof of the Discrete Hexachordal Theorem. Instead, we show below
that the freedom to use any integrable weight function renders the Discrete
Hexachordal Theorem 1 an immediate corollary of the Continuous Hexachordal
Theorem 2.

2.4 Discrete Theorem as Corollary

Suppose a discrete rhythm R has weights (w0 , w1 , . . . , wn−1 ), with each weight
either 1 or 0. Then define the step function f (x) = wi for ni ≤ x < i+1 n . For
example, Fig. 3(a) shows the step function corresponding to the top rhythm
in Fig. 1, whose discrete weights are (1, 1, 0, 0, 1, 0, 0, 1). Note that the total
weight/area is 4· 18 = 12 , which accords with the discrete weight of 12 n= 12 8=4.
We formalize this correspondence between continuous and discrete as follows:

Corollary 1. The Discrete Hexachordal Theorem 1 follows from the Continuous

Hexachordal Theorem 2.

Proof. We use the notation

18 B. Ballinger et al.


1, for all x ∈ A
χA (x) =
0, otherwise
to represent the 1/0 characteristic function of a set A.
We convert the discrete rhythm (w0 , w1 , . . . , wn−1 ) into the continuous rhythm


n−1 
f (x) = wi · χ[ i , i+1 ) .
n n
i=0

This has the feature, mentioned above, that for all x ∈ ni , i+1
n , we have f (x) = wi .
Because of the horizontal compression
n−1 involved in this conversion, the discrete
histogram contribution HR (d) = i=0 wi wi+d corresponds to the continuous
histogram contribution
   1  
d d
HR = f (x)f x + dx
n 0 n
 1 n−1
 
  
d
= wi · χ[ i , i+1 ) f x + dx
0 i=0
n n n
  1
n−1 
d
 
= wi χ[ i , i+1 ) · f x + dx
i=0 0 n n n
 i+1   

n−1
n d
= wi f x+ dx
i=0 n
i n
 i+d+1 

n−1
n
= wi f (x) dx
i+d
i=0 n
 

n−1 i+d+1
n
= wi wi+d dx
i+d
i=0 n

1 
n−1
= wi wi+d
n i=0

So, the continuous histogram is proportional to the discrete histogram at integral

values of d (see Fig. 3(b)), and the conclusion of the Continuous Hexachordal
Theorem 2 that R is homometric to R implies the same in the discrete case,
which is precisely the claim of the Discrete Hexachordal Theorem 1.

2.5 Double-Counting Diameter Intervals

We return to the the issue of double-counting an interval that equals the di-
ameter (d = n/2 in the discrete case or d = 12 in the continuous case) in the
histogram HR (d). In music the diameter in the case of an equal-temperament
 The Continuous Hexachordal Theorem 19

scale corresponds to a tritone. Recall from Definition 2 that the continuous his-
1
togram is defined by the equation HR (d) = 0 f (x)f (x + d) dx. Applying this
for d = 12 to the step function f (x) in Figure 3 results in
 1
1 1
HR ( ) = f (x)f (x + ) dx.
2 0 2

When x ∈ [0, 18 ), the product f (x)f (x + 12 ) is 1. And also when x ∈ [ 12 , 58 ),

the product is again 1, because x + 12 wraps around to [0, 18 ). For all other x,
the product is 0. So HR ( 12 ) = 2 · 18 = 14 , which corresponds to the height 2 for
d = 4 in the discrete case in Figure 1. Thus, the continuous histogram analog
also “double-counts” the diameter d = 12 .
Moreover, we can see that this is the natural definition, by considering d =
1
2 − ε for some small ε > 0. The same integral leads to HR ( 12 − ε) = 2( 18 − ε)
which goes to 14 as ε → 0. Thus, the height HR ( 12 ) is consistent with the limit
for d < 12 . Stipulating that d = 12 should be treated specially would destroy this
natural correspondence.

2.6 Patterson’s First Theorem

Patterson’s first Theorem [Pat44] goes beyond the k = n/2 precondition of the
Discrete Hexachordal Theorem 1. It may be stated as: two homometric (n, k)-
rhythms have homometric complements. In our continuous generalizations, two
rhythms with the same number k of onsets have the same weight. So the gener-
alization is:

Theorem 3. If R1 and R2 are two integrable rhythms on the continuous cir-

cle with equal weights, W (R1 ) = W (R2 ), and they are homometric, i.e., for
all distances d, HR1 (d) = HR2 (d), then their complements are homometric:
HR1 (d) = HR2 (d).

Proof. Let the weight function of R1 be f (x) and that of R2 be g(x). Fix a dis-
tance d. We compute HR1 (d) and show it is equal to HR2 (d). From Definitions 2
and 1, we have
 1
HR1 (d) = f (x) f (x + d) dx
0
 1
= (1 − f (x))(1 − f (x + d)) dx
0

Multiplying out terms and separating integrals yields

 1  1  1
= 1 dx − 2 f (x) dx + f (x)f (x + d) dx
0 0 0
 1  1
=1−2 f (x) dx + f (x)f (x + d) dx
0 0
20 B. Ballinger et al.
1 1
Now, because W (R1 ) = W (R2 ), we have 0 f (x)dx = 0 g(x)dx, and because
1 1
R1 and R2 are homometric, we have 0 f (x)f (x + d)dx = 0 g(x)g(x + d)dx:
 1  1
=1−2 g(x) dx + g(x)g(x + d) dx
0 0

However, we know, by the same reasoning, that this expression is

 1
= g(x) g(x + c) dx
0

And we have therefore established that the complementary rhythms are homo-
metric:
 1  1
f (x) f (x + d) dx = g(x) g(x + d) dx
0 0
HR1 (d) = HR2 (d)

3 Open Problems
Our results may be interpreted in terms of polyphonic rhythms, in which several
instruments are linearly combined [OTT08]. For instance, to model three identi-
cal drums playing together, interpret the weight f (x) = 13 to mean that one drum
is struck on a particular beat, while the weight f (x) = 1 would mean all three
are struck. It would be interesting to explore whether homometric polyphonic
rhythms have a musical significance.
We know that two sets of points with different cardinalities and different
weights may be homometric, but we neither understand the constraints here
mathematically nor know if there is any musical interpretation of such sets.
Theorem 2 generalizes to weights in [0, 1] on a sphere, with distances measured
by geodesics, and with W (R) = 12 corresponding to the integral over a hemi-
sphere equalling 12 . The discrete analog is “distance regular” points on a sphere,
e.g., the vertices of a Platonic solid. Is there any musical analog for spheres in
any dimension?

Acknowledgements
The authors like to thank the anonymous referees for their useful comments.

References
[AG00] Althuis, T.A., Göbel, F.: Z-related pairs in microtonal systems. Memorandum
1524, University of Twente, The Netherlands (April 2000)
[Ami07] Amiot, E.: David Lewin and maximally even sets. Journal of Mathematics
and Music 1(3), 157–172 (2007)
[Bla99] Blau, S.K.: The hexachordal theorem: A mathematical look at interval re-
lations in twelve-tone composition. Mathematics Magazine 72(4), 310–313
(1999)
 The Continuous Hexachordal Theorem 21

[Bue76] Buerger, M.J.: Proofs and generalizations of Patterson’s theorems on homo-

metric complementary sets. Zeitschrift für Kristallographie 143, 79–98 (1975)
[Bue78] Buerger, M.J.: Interpoint distances in cyclotomic sets. The Canadian Miner-
alogist 16, 301–314 (1978)
[For77] Forte, A.: The Structure of Atonal Music. The Yale University Press, Madison
(1977)
[Igl81] Iglesias, J.E.: On Patterson’s cyclotomic sets and how to count them.
Zeitschrift für Kristallographie 156, 187–196 (1981)
[Jed06] Jedrzejewski, G.: Mathematical Theory of Music. Editions Delatour France
(2006)
[JK06] Jaming, P., Kolountzakis, M.: Reconstruction of functions from their triple
correlations. New York Journal of Mathematics 9, 149–164 (2003)
[Joh03] Johnson, T.: Foundations of Diatonic Theory. Key College Publishing (2003)
[Lew59] Lewin, D.: Intervallic relations betwen two collections of notes. Journal of
Music Theory 3(2), 298–301 (1959)
[Lew60] Lewin, D.: The intervallic content of a collection of notes, intervallic relations
between a collection of notes and its complement: An application to schoen-
berg’s hexachordal pieces. Journal of Music Theory 4(1), 98–101 (1960)
[Lew76] Lewin, D.: On the interval content of invertible hexachords. Journal of Music
Theory 20(2), 185–188 (Autumn 1976)
[Lew87] Lewin, D.: Generalized Musical Intervals and Transformations. Yale Univer-
sity Press (1987)
[Maz03] Mazzola, G.: The Topos of Music. Birkhäuser, Basel (2003)
[Mor90] Morris, R.D.: Pitch-class complementation and its generalizations. Journal of
Music Theory 34(2), 175–245 (Autumn 1990)
[OTT08] O’Rourke, J., Taslakian, P., Toussaint, G.: A pumping lemma for homometric
rhythms. In: Proc. 20th Canad. Conf. Comput. Geom., August 2008, pp. 99–
102 (2008)
[Pat44] Lindo Patterson, A.: Ambiguities in the x-ray analysis of crystal structures.
Physical Review 64(5-6), 195–201 (1944)
[Reg74] Regener, E.: On Allen Forte’s theory of chords. Perspectives of New Mu-
sic 13(1), 191–212 (Autumn-Winter 1974)
[Sen08] Senechal, M.: A point set puzzle revisited. European Journal of Combina-
torics 29(1), 1933–1944 (2008)
[Sod95] Soderberg, S.: Z-related sets as dual inversions. Journal of Music The-
ory 39(1), 77–100 (Spring 1995)
[Tou05] Toussaint, G.T.: The geometry of musical rhythm. In: Akiyama, J., Kano,
M., Tan, X. (eds.) JCDCG 2004. LNCS, vol. 3742, pp. 198–212. Springer,
Heidelberg (2005)
[Tym06] Tymoczko, D.: The geometry of musical chords. Science 313(72), 72–74 (2006)
[Vuz85] Vuza, D.: Sur le rythme périodique. Revue Roumaine de Linguistique -
Cahiers de Linguistique Théorique et Appliquée 22(1), 103173–103188 (1985)
[Wri78] Wright, O.: The Modal System of Arab and Persian Music AD 1250-1300.
Oxford University Press, Oxford (1978)