Clough - Maximally Even Sets
Clough - Maximally Even Sets
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=duke.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].
Duke University Press and Yale University Department of Music are collaborating with JSTOR to digitize,
preserve and extend access to Journal of Music Theory.
http://www.jstor.org
MAXIMALLY EVEN SETS
93
the scales treated by Joseph Yasser in his monumental study are re-
ducible to ME sets. But, as far as we know, ME sets have not been
previously recognized and studied as a class, either within the 12-note
universe or in general.
Part 1 of the paper is devoted to matters of definition and construc-
tion, questions of existence and uniqueness, combinatorialproperties,
and interval content of ME sets. In part 2 we view the usual diatonic
set as a member of a special family of ME sets found in chromatic
universes of certain sizes, and show that the set of all such "diatonic"
sets is the set of ME sets that have exactly one ambiguity. The third
and final part deals with generated ME sets, complements, maximally
even circles of intervals, and second-order ME sets.
94
Unless otherwise stipulated, we assume that the elements of any set
of pcs are represented by smallest non-negative integers (mod c):
0, 1, ,c-1.
Example 1.1. Let D12.7 = {0, 2, 4, 5, 7, 9, 11}, with (pitch class)
C=O. Here D12.7 represents the represents the C major diatonic set.
The symbol D(12,7) represents the set of all seven-note subsets of the
twelve-note chromatic universe.
Definition 1.3. Let D(.,j = {D, DI, D1, ... D<,-1}. Then
95
for any s in S there exists an N, 0 - N - d - 1, such that
D(N+I) - DN=S (mod c), where subscripts are reduced (mod d).
Example 1.3. In the C major set, <1> = {1,2}, <2> =
{3,4}, < 3 > = {5,6}, < 4 > = {6,7}, etc. Without ambiguitywe may
also write, for example, < C, E > = < 2 > or < C, E> = {3,4}.
Note that in any set < 0 > = 0. The fact that some dlens correspond
to more than one clen is related to the notion of "quality"(i.e., major,
minor, etc., as in the scalar intervals of traditional harmony).
ME set defined
96
we use the symbol M,,t with no specific assigned value, to mean "any
member of M(c,d)."
Example 1.4. In the octatonic set Dl28 = {0, 1, 3, 4, 6, 7, 9, 10},
we find the following interval spectrums: < 0 > = {0}, < 1 > = {1,2},
<2> ={3}, <3>={4,5}, <4>={6}, <5>={7,8}, < 6 > = {9},
and < 7 > = {10,11}. Since each of these spectrumsis a single integer
or two consecutive integers, D1.,s is a ME set, and we may write
M12.8 = {0, 1, 3, 4, 6, 7, 9, 10}. (It follows easily that all octatonic sets
are ME; later we will prove the general principle involved in this as-
sertion.) The symbol M(12,8) represents the set of all ME sets with
c = 12, d = 8. As we will see, this is precisely the set of all octatonic
sets.
Here is a method for constructing ME sets, which may provide a
more concrete idea of what it means to distribute notes "as evenly as
possible": First, choose values for c and d, say 12 and 7, respectively;
locate d "white" points equidistantly around the circumference of a
circle; and do the same for c-d (in this case 5) "black" points on an-
other circle. These two steps are shown in fig. 1, a and b. Next su-
perimpose one circle on the other so that no two points are in the same
location, as in fig. Ic. Finally, assign dlen 1 to all adjacent pairs of
white points, and clen 1 to all adjacent pairs of points regardless of
color, and "tune" the system as desired-to equal temperament or
whatever-as shown in fig. ld. The white points now represent a ME
set with parameters c and d, in this case M12.7, and the black points
represent a complementaryME set with parametersc and c-d, here
M12,5.(As we shall see, complements of ME sets are themselves max-
imally even.) A proof that the construction described here generates
ME sets will be given in part 3 of the paper.
The greatest common divisor of c and d is notated "(c,d)." We es-
tablish three classes of ME sets based on (c,d): class A, where
(c,d) = d; class B, where 1 < (c,d) < d; and class C, where (c,d) = 1.
Familiarsets appear as representativesof each class, as shown in figs.
2a, 2b, and 2c. For comparison, fig. 2d is a familiarset that is not max-
imally even: the set of the melodic minor ascending scale has three
sizes of dlen 3 (the traditional4th)-one too many for maximaleven-
ness; that is because its half steps do not have maximalseparation.The
significance of this classification will become clear as we discuss the
various properties of ME sets.
97
(a) 7-note universe (b) 5-note universe
98
(a) whole-tone set: c = 12, d = 6; (b) octatonic set: c = 12, d = 8:
(c,d) = d 1 <(r(.)=4<d
Dc,d= {[Oc]rI...
[c] [2c] (d- 1)c]}
j}-l d
Then all nonzero interval spectrums of Dc,d are sets of two consecutive
integers. Therefore Dc,d E M(c,d).
99
Ordinarily,the transpositionsof a pcset are identified by the trans-
position operators. For example, the transpositions of X = {O, 1, 2}
might be labeled as follows: X) = {0, 1, 2}, Xl = {1, 2, 3}, X2 =
{2, 3, 4}, etc. However, our work requires that the transpositionsbe
labeled in "cyclic"order. We will set up the mechanismsfor this now,
and explain what we mean by "cyclic" as the exposition proceeds.
Definition1.8. Let c,d,m, N be integers such that 0 in < c- 1 and
0 - N < d - c. Then the J-functionwith these parametersis expressed
by
cN + in
JJC. (N)
i? =
[?.c~ d
Definition 1.9. Let c, d, in be as in the above definition. Then the
J-set with these parameters is the set
J. =, { c. (0), J( (1) . . .,
(d-1)}.
The Clough-Myerson algorithm for generating "diatonic" sets,
quoted in theorem 1.1 above, is the basis for our J-set.
Jc. (N) is the (N + 1)th pc of the J-set J"', ; that is, J"', (0) is the first
pc, J',,t (1) is the second pc, etc. The parametersc and d are the chro-
matic and diatonic cardinalities. The superscriptm, called the mode
index, is a label for a particular transposition; however, setting in
equal to 0, 1, 2, . . ., etc. does not generate the transpositions of
J'", in the "expected" order. Instead it generates them in an order
correspondingto their internal cyclic structure. For example,
2.7= {7} +
J2.7
12,7={7}+ 12.7,
100
This order of generation corresponds to the internal structure of
J17, which is also based on clen 7. For example, J127 {0-7,
1.7, 27 . ..,6.7}{, 7, 2, 9, 4, 11, 6} = 0, 2, 4, 6, 7, 9, 11}
(mod 12).
For ME sets that do not have a "circle of fifths,"-those with
(c, d) 1-the situation is somewhat different, but still correspondsto
a cyclic aspect of the set. We comment further on this point below in
connection with the case (c,d) 1, and we take up the question of gen-
erators in detail in part 3.
Note that J' .7 is the "normal order" of the major scale set as de-
fined by Forte (1973, 3-4). In general, the Clough-Myersonalgorithm
generates sets in normal order. We leave it to the reader to discover
the reason for this.
The following lemma and corollaries pertain to the interval of I
steps in a J-set-the interval from the (N+ 1)th to the (N+I+ l)th
pc-from J'" (N) to J' (N + I), where the expressions in parentheses
are appropriatelyreduced (mod d). This is the interval of dlen I.
Lemma 1.1. Let c,d,mn,N,I be integers such that 0 < N < d - c, and
1 II d-1. Let cN + m = dqN + rN and cl = dql + rl, where
qN, ql, rv, and r/ are integers and 0 < rN < d and 0 c r/ < d. Then
jJ. Id (N "I M M
+I,,)a -- J ,(N)- 'JCd(N + I) - J1,d (N)
N +
[c(N + I) + in
-- d \
cN+ in (mod c).
- c(N +)+m
d ] \
-cN + in (mod
_[Ndm c).
101
In either case
(in _+ in -c(N +I) + m 1 cN+m]
Jcd (N + I)d- c,d(N) c d [ d
[ [rN d+ ](modc).e
Corollary 1.1.
103
Proof. Since (c,d) = 1, there are no single-element spectrumsex-
cept < 0 > = {0} (lemma 1.3). Thus all M,. must have both CP and
MP. It follows that they must be rounded. Since these sets have MP,
they must have CV (lemma 1.4). We have now established that (for c,
d as stated) all M,.1 are rounded and have CV. It follows that they are
equivalent under transposition (lemma 1.5)..
We now prove one more lemma that will enable us to show that all
ME sets with (c,d) = 1 have J-representations.
Lemma 1.6. Let 1 - d < c, 0 < in c - 1, and j be any non-
negative integer. Then
J..1+i ={1} + J,.,
where the subscriptc following the parenthesizedexpression indicates
reduction (mod c).
Proof. Let j = 1 and suppose in + d < c. Then
Jt,+l (N) = l ' (N)
cN + (n +d)
-L d \
[CN+ n1
It follows that
{
j, (,,+,i, = l+jm
,r+c(N-l)+in
=- 1+L
= I + J..",(N- 1)(mod c).
IfN = 0,then
+ c
J'(1+ (0) [n d ]
n-c
=1+ d
n-c
-1+ +c
d0
104
c(d-
=1+ dl)+im]
1+ J',t (d-l) (mod c).
Thus
J (/? d), I=+ {1} +
J,t J.lt
Then by induction we see that
Jcldl, = {} + J , / .
Theorem 1.4. For any fixed c and d, (c,d) = 1, and any M(./, there
existsanin, 0 < m < c- 1, such that M,., = J''.t
Proof. We know that for any fixed c and d, (c,d)= 1, all M,., are
equivalent under transposition (theorem 1.3). Also J'., is a ME set
(theorem 1.2). Thus M.,l = {/} + J,, for some integerj. It follows that
M.d = Jl,,, where m = jd (mod c) (lemma 1.6). Further, since m is re-
duced (mod c), we have 0 -< n < c - 1.e
Now we are ready to relax the condition (c,d) = 1. We will show
that M,.Xlwith (c,d) = k # 1 is "generated"by M,. ,, with (c',d') = 1.
To get a preliminaryidea of what "generated"means in this case, see
fig. 3, which shows how the ME set J7. 9 is generated by the smaller
ME set J.3.
To see how this structure is reflected in the definition of the J-set,
consider J7 .9, as m is set to 0,1,2, etc.:
J'l9 = {0, 1, 3, 5, 6, 8, 10, 11, 13}, in = 0, 1, 2;
Jl ' = {0, 2, 3, 5, 7, 8, 10, 12, 13}, in = 3, 4, 5;
105
1101+J5,3 2
0?o (mod 15) \
[ 5'3 *15.9 4 3
~2 10 nn = 67.8X
j2 =
5,3 = 0,2,41}
151 +J.3 (mod 15)
y7 j2 yj
J15,9 J53 u |(5) +J5.3j (rmod15)
there are d intervals of dlen jdlk, if one interval has a clen less than (or
greater than) jc/k there must be at least one interval whose clen is
greater than (or less than) jc/k, contradicting the assumption that
M,(, is a ME set. Thus < jd/k > = {jc/k}.
Lemma 1.8. Let M( ( = {Do, D, . D. , D(_} and suppose
(c, d) = k. Let c' = clk and d' = d/k, M, =
c'= c/k and d'=d/k, Mc',d'={DO, D .. ., D,i_1}, and M, =
M,(d=UtM,.
Proof. We know that < jd/k > = {jc/k} (lemma 1.7). Thus Dj, =
jc' + Do, Di+j/, = jc' + Di, . . . D,l-l+(j = jc' + D,-l, where
0 < j < k - 1. It follows that M,.,/ {D,, D , . . , D/ -_,c' +
Do ..., c' + D,- ...
..., (k - 1)c' + D, ... (k -
1)c' + D,_ = U1M/ . ?
We see that M,,,i is completely determined by M,.,l. The notation
MC,1,suggests that M, ,, is a ME set with chromatic and diatonic car-
dinalities c' and d'. This is fact the case, but we must enter one more
lemma before we prove it.
106
Lemma 1.9. For any Dc,d and any associated dlen I, if < I =
{kl or {kl, kl + 1}, then
cI
k, =
kl H *-
Proof. Suppose there are h intervals of dlen I such that
DN+I,, - DN k1(mod c),O - N - d - 1. Then there are d - h in-
tervals of dlen I such that D(N+I,, - DN = k + 1 (mod c). Thus the sum
of the dens of these intervals is hk1 + (d - h)(k1+ 1). But this sum is
also cl (lemma 1.2). Thus
hk, + (d-h)(k + 1) = cl
cI d-h
7V
cl d-h d-h
since - - < 1.?
d is an integer and 0
Lemma 1.10. Let Mcd = {Do, D1, . . .,Dd-} and suppose
c' = c/(c,d), d' = d/(c,d) and Dc,d' = {Do, D1.. , Dd' }. Then
Dc',d' is an element of M(c',d').
Proof. We must show
f] nc7l
nc,
- nc' + 1 (mod c).
But since c' divides c we have
107
We are now in a position to prove the generalizationsof theorems
1.3 and 1.4.
Theorem 1.5. For any M,., there exists an in, 0 - in c - 1, such
that M,,I = J-.
Proof. Let (c,d) = k, c' = c/k, and d' = d/k. Then M(,.,, as in
lemma 1.10 is a ME set. Since (c',') = 1 there exists an in', 0
in' < c' - I such that MA,.,,/= Ji'! (theorem 1.4). It follows that
M,.</= UkM, where MA= {jc'} + J"' (lemma 1.8). Now let m
kin'. Clearly 0 mn< c- 1. Also J'" (N + d'j) = c' + J. (N)
whenO j -< < k landO < N < d' - 1. Thus
-
U (i=jc} +
J7.d= - 1'.,r)
= Uk-I,m
U=^Mi
= M ../ -
Theorem 1.6. If D(,, = {j} + M,.(,,then D,.., EM(c,d); that is, any
transpositionof a ME set is a ME set.
Proof. If D(.,( is a transpositionof a ME set, then
D, = {j) + M,.,t
= {/} + Ji",
for some integer in, 0 - in < c - 1 (theorem 1.5). Then
M,C.,= U i= MiA
and
M.,, =U =,,
where MJ = i{c'} + J"'l, (mod c) and M {c' + "' (mod c)
(lemmas 1.8 and 1.10, and theorem 1.4).
Therefore
108
{i} + M] .
It follows that
M,.d =U , MM;
-
U ({i} + M)
={i}+ U.-,l
= {i} + M,,.
+mt r [Oc+m]c
d2c +
[lcd ,c d ].
-
c d
- {j}
cd- n c(d- 1)-in
+ d d
109
c 2- n c 1- m
d ' d
- i J
fc .0-n c1 - n
(i.e. d [ d (mod)
= {d} +1 -
110
Proof. We need to show that correspondingelements of these sets
are equal. Choose any N, 0 - N < d - 1, and i, 0 < i - (c,d) -
1. Then
Jic(t.I+i (N) =
N +j(c,d) + i]
Also
[cN+ (i + )(c,)-1 [c'N +j + ((c,) - l)/c,)
r+(c,
d))- 1 )
111
<
It follows that
[r + ((c,d)- )/(c,d)
q+ d' j=<=q+O=q.
Hence Jlt''','+ (N) = q for all i, 0 < i < (c,d) - 1. Since Nwaschosen
arbitrarily,we conclude that
jlcI(t_ =_
jijc.dl - Jjlc. i .l+c.
+ dl
( 1I
ca.= c.d ?- .
c.d
Cc
Corollary 1.3. I M(c,d) I (
-
Proof. Let j be as in lemma 1.11. Then, for each j, 0 j <
c
c7-T - 1, there are (c,d) equivalent J-sets, each with a distinct mode
indexrn, 0 - m < c - 1. Further, fori, 0 - i (c,d) - 1,we have
c-l
112
Lemma 1.14. Let mi and m2 be integers such that 0 < mi < m2 '
_Then I
'T"I,., ylm (c,di) _ jml2(C,d)
(c,) -1. jncd) jnC,d
c d
Proof. Let c' = (cd and d' = ( - Then 0< m2 - mi c' -
1,
fc+ml Mi
j,mn~c,M)
c,d [_ ' [2c+ d m-c'+m d '+
] ] ]}
and
jm,c,d)
i c,d {T[7
[c'+ 2] [2c'+ 2] [ (d- )c'+m
[j d' d' d'
where the elements of the sets are arranged in order of magnitude
(mod c) (lemma 1.13). If any of the corresponding elements differ,
then the sets cannot be equal. Now there exist integers N and
q, 0 N< d',q - 0, such that c'N - d'(q + 1) = - m2 (lemma
1.12). Thusc'N = d'q + (d' - m2). Then
[c'N+ ] = [d'q + (d'- m2) + m2
d [ d'-d
= [c l] + 1.
d+
Thus Jmc,d) (N) > Jnc,d) (N) + 1, and we have found two corre-
sponding elements that differ by at least 1. It follows that
Jm{(c,d) A jm2(c,d).
c,d c,d
C
Corollary1.4. I M(c,d) I- c--d .
C
Proof. Lemma 1.14 implies that for 0 i,j < c,--
1, cJ'i'd
,d = jj(c,d if and only if i = j. Hence there must be at least
C
(c,) distinct pcsets in M(c,d).?
c
Theorem 1.9. For fixed c,d, I M(c,d) I = -c,-d' and
M(c,d) = {J?d, (c,d), 2(c,d)
c,d) , ,
c,d
-d
c,d I
113
Proof. The fact that there are C-- distinctpcsets in M(c,d) follows
from corollaries 1.3 and 1.4. Their J-representation follows from
lemma 1.14. 0
Counting problems like the above will be familiar to students of
atonal set theory, where the distinct pcsets generated by a specified
group of operations are often called the "forms" of a pcset. Thus, if
the group of operations includes (i) transposition and (ii) inversion
about a fixed axis followed by transposition, there are three forms of
the octatonic set. If we restrict the group of operations to (i) above,
there are still three forms, since inversionsof this particularset are not
distinct from transpositions. Following John Rahn (1980, 90-91), we
call the number of distinct operations (within the specified group) that
map a set into itself the "degree of symmetry."Dividing the degree of
symmetry of a set into the number of operations in the group then
yields the number of distinct pcsets corresponding to the set. Since
ME sets are inversionallysymmetrical,it makes no difference whether
we choose (i), or (i) and (ii) above as the group of operations; the
number of distinct pcsets corresponding to a particular Mc,dis the
same in either case.
We now turn to the second question raised above. Let us first re-
state it in more precise terms: Suppose we have before us all of the
ME sets for some given c and d. We know from theorem 1.9 that there
c
are (c-,d such sets. Now we select a particularpc and ask, how many
of these sets include the selected pc? Theorem 1.10 provides the an-
swer, but first we need to have some terminology to discuss "pcsets in
M(c,d) which include p." We call the selected pc a "tonic," and we
make the following definition:
Definition 1.16. A tonic ME (TME) set ,d is an M,d which in-
cludes a specified pitch-classp. We write TP(c,d)to represent the set
of all ME sets with parameters c,d that contain the pitch-classp.
As in the case of Mc,d, the symbol ,d is the name for a variable
whose assigned value is always a pcset of certain description. As an
example of TME sets, the five sets correspondingto TO5 are shown in
fig. 4a. All of these sets are members of M(8,5) of course. There are
three additional sets in M(8,5) which do not include p = 0; these are
shown in fig. 4b.
d
Lemma 1.15. For any fixed c,d, I T? (c,d) I = (c-,' and
TO(c,d) =J ,d J (cd) j2(c,d)
* * * Jd-c,d)}
c,d c,d , d c,d
114
(a) Five sets correspondingtoT(
d
(c,d) and
0 if(0) = ifid-
0 - 1.
c,=and= only
/O(c,d)
r(c,d) {jojIc,dt i2(c,d c,d)} . I
Thus
Tc,(c,
}
{y,,J,,d ^, c,d
d
Theorem 1.10. For any fixed c,d and any pc p, I 7P(c,d)I = 7,d
and
P(c,d)
i \L,U = \{cjpdc
c,d
j((c,d}+pdc
a c,d
,j (2(c,d)+pd)d(c,d+pd
Jc,
jd-c,d)}pd
Jc,d
116
Do, D1, . . .Dd-1;
Interval content
117
Proof. From lemma 1.9 we know that the clen of any interval in
<I> is ]
or [] + 1. Let h be the numberof intervalsin < I > of
clen [X] + 1. Then there will be d - h intervals in < I > of clen
[E]. The sum of these dens is
= dq + r-d[ ]
- r.
Thus there are r intervalsin < I > of clen [ + 1 and hence d-r in-
tervals in < I > of clen [] . From lemma 1.16 it follows immediately
that there are r intervals in < d - I> of clen c - ([ + 1) and
d - r intervals in < d - I> of clen c - [ .
118
plete profile provided by DFUNC seems potentially advantageous.
For example, DFUNC is sufficient to distinguish between any two
Z-related sets in the 12-pc universe; whether this holds for Z-related
pairs (and triples, etc.) in any chromaticuniverse is an open question.
DFUNC (M85 0,0) = 5
DFUNC(M8. 11,)= 2--------
DFUNC (M8.5 2,1)= 3---------
DFUNC(M.5 3,2) =4-.---
DFUNC (MX85 4,2)= 1--
DFUNC (M85 4,3)= 1..1
I
DFUNC (MX5 ,3) =
DFUNC(M8.
DFUNC
5,3) =4------
6,4)= 3----------
(M8.5.
DFUNC (M8.5 7.4) = 2- ----------
Figure 5. The interval profile of M8,5= {0,1, 3, 4, 6}
119
We now set forth a few additional concepts which will allow us to
show how the spectrum of a set is associated with other properties of
the set.
Definition 1.19. Cc = {0, 1, 2, ..., c - 1}, a multiset.
Note that while Uc (the chromatic universe) and Cc contain the
same elements, they are not identical. Ucis a set, whereas Ccis a mul-
tiset in which each element has multiplicity one.
Definition 1.20. If < 1 > = {1, 2}, we say that the set is reduced.
The above definition is borrowed from Clough and Myerson.
Lemma 1.17. Suppose c and d are positive integers such that
< d -< c. Let S = Ud , FI' + 1 be a set (not a multiset)
I=1J]
where the elements are reduced (mod c). Then S = Uc=
{0, 1, 2, .... c - 1}.
Proof. The proof is in three parts:
(i) 0 = [] (trivial).
So if
fc/1 (I + 1) - 1) .
[] + 2[ d , and \c{d + 1-c-1,
then all the integers from 0 to c-1, inclusive, will also be in S. Thus
S = {0, 1,2 ... c- 1}.
c(c- 1)
Lc-.
= c-l.
120
If c< d<c, then when I = d - wehave
dc - d - (cd)]1d) +
=
"[' d
-dLd j
=.+[ (c- d)]
1
=c-1 (since c< d < c).
In either case c - 1 E S.
(iii)
+ (c d)]-
[]clI +2,~[cI [c+(cd)]+ 1 (since d>)
=- d 2
-cI+
1)].
[ +1 q +1
121
<
\c-d 1
= q + 1 + [-] (since c < d
< c)
_qd+c]
dL+c1
CI+C
122
Closely related to our efforts here are Norman Carey and David
Clampitt's(1989) work and Eytan Agmon's (1989) work. In this part
of the paper and in the third and final part, we seek to advance the
inquiryinto the essence of diatonicism. We will not attempt to review
all of the importantwork cited above, but we will point out the more
salient connections, particularly those between Agmon's work and
ours.
Clough and Myerson show that the usual diatonic set is one of an
infinite class of "special" sets with the following properties: (1) "car-
dinality equals variety"-a "line class" including n different pcs
appear in n species, (2) "partitioning"-chord species map (unambig-
uously) onto chord genera, and (3) the "deep-set" property as studied
by Gamer (1967, 32-59). They offer a definition of this class of sets
based on Myhill's property together with the "generalized circle of
fifths," and show that, in terms of the relationship between c and d,
there are just two families of "special" sets: family A, where c =
2d - 1, and family B, where c = 2d - 2 with d odd. M8,5and
M12,7, discussed above and shown in figs. 4 and 2c, respectively, are
members of family B. For comparison, fig. 6 shows the two ME sets
of family A with d = 5 and d = 7: M9,5 and M13,7.
Agmon (1989) offers a mathematicalapproachto the same class of
"special" sets-an approach closely resembling that of Clough-
Myerson in some respects and differingfrom it significantlyin others.
Agmon measures intervals by clen and dlen, as do Clough and My-
erson, and as we do here. He distinguishesbetween the sets of family
A, and those of family B, as defined above and in Clough and Myer-
son. In defining the class diatonic, Agmon quite reasonably rules out
the sets of family A on the ground that they are inherently less inter-
esting than those of family B, because the former are generated by the
whole-step (clen 2) or its complement, while (for c > 4) the latter are
generated by a "skip" of more than one diatonic step (e.g., in the
usual diatonic set, clen 5 or 7-the perfect 4th or perfect 5th). The
question of whether the sets of family A are less interesting has a sur-
prisingtwist, which we come to in part 3 of the paper. Meanwhile, we
accept Agmon's premise and focus on the usual diatonic and other sets
of family B.
Some new terminologywill be useful in maintainingthe distinctions
among various kinds of sets. We will call the sets of family B (ME sets
with c = 2d - 2 and d odd) diatonicsets. We will continue to refer to
M12,7 as the usual diatonic set, and we will henceforth refer to other
diatonic sets as hyperdiatonicsets. Thus our diatonic sets, which in-
clude the usual diatonic set plus all hyperdiatonicsets, correspond to
Agmon's "diatonic systems." As before, we will refer to the comple-
ment of M12,7, M12,5as the usual pentatonic set.
123
Figure 6. Examples of family A: c = 2d - 1
124
Ambiguities
We first define some terms pertaining to correspondences among
dlens and dens. Our definitions are those given by Jay Rahn in an
article elsewhere in this journal, reworded to suit the context of our
discussion. Our method of counting the number of cases of difference
or ambiguity or contradiction in a set also follows that of Rahn.
Definition 2.1. Given Dp, Dq, D,., Ds E DCd, if dlen(Dp, Dq) =
dlen(D,, Ds), and clen(Dp, Dq) = clen(D,, Ds), that is a case of differ-
ence.
Definition 2.2. Given Dp, Dq, D,, Ds E Dc,d, if dlen(Dp, Dq) =
dlen (D,., Ds), and clen(Dp, Dq) = clen(D,, Ds), that is a case of am-
biguity.
Definition 2.3. Given Dp, Dq, D,., Ds E Dc,d, if dlen(Dp, Dq) <
dlen(D,., Ds), and clen(Dp, Dq) > clen(D,., Ds), that is a case of con-
tradiction.
Examples 2.1. Suppose D,d = {C, D, Eb, F#, G, A, B} =
{0, 2, 3, 6, 7, 8, 11}. Then
(1) dlen(C, Eb) = dlen(D, Fg) and clen(C, Eb) dlen(D, F#)-a
case of difference;
(2) dlen(C, Ft) $ dlen(Ft C,) and clen(C, Ft) = clen(Ft, C)-a
case of ambiguity;
(3) dlen(E,, F#) < dlen(F#, Ab) and clen(Eb, F#) > clen(F#, Ab)
-a case of contradiction.
Note that each case of difference, ambiguity, or contradiction in-
volves two intervals-that is, two ordered pairs of pcs. The two in-
tervals may have zero, one, or two pcs in common, though in different
order positions (all three possibilities are evident in the examples
above). It follows immediately from the above definitions and from
lemma 1.16 that for each case of difference, ambiguity, or contradic-
tion involving particulardlens and dens, there is a correspondingcase
involving the complementary dlens and dens.
Examples2.2. In M12,7,< 2 > = {3, 4}, implyingone or more cases
of difference. (There are in fact 12 = 3 .4 cases of difference here--12
pairs of intervals-since there are three intervals of dlen 2 with clen
4[major thirds], and four intervals of dlen 2 with clen 3 [minor
thirds].) The complementary cases arise from <7 - 2> = <5>
= {12 - 3, 12 - 4} = {8, 9}. Complementary cases are not necessar-
ily distinct. For example in M12,7, < 3 > = {5, 6} and < 4 > =
{6, 7}, implyingone or more cases of ambiguity. There is in fact a sin-
gle ambiguityhere-that is, a single ambiguouspair of intervals-one
interval of dlen 3 with clen 6, and one interval of dlen 4 with clen 6.
But these two intervals are mutually complementary.
125
We now state a number of results connecting the above definitions
with ME sets.
Lemma 2.1. No Mc,d contains a contradiction.
Proof. By theorem. 1.11, for any dlen I,
Proof. If d 22 then
c(+l) [] = [c I + c] [c
- +
2]- [c]
=2.
c0O
- c-1 c 2 c(d-1)
<d < d < d '< d
126
As the following two lemmas show, any ambiguityin a ME set in-
volves two consecutive diens, each with a two-element spectrum.
Lemma 2.4. If co is an element in two distinct spectrums of Mc,d
(i.e., co is the clen of an ambiguity), then there exists a unique
I, 0< I d- lsuchthat < I > < I + 1> = {co}.Further
COW
+1=[
[]
+ 1 = [c(+d1)]
=
d d
Proof. By lemma 1.9, co must have the form [] or [J] +
dq - c] 1
and
[ 1
This implies d - c. By definition d ' c. Thus d = c. But if d =
c, then (c,d) = d, which implies Mc,dhas no ambiguity (lemma 2.3),
contradictingthe assumption <I> n <I + 1 > 7 0. Thus < I +
1 > must be a two-element spectrum.?
Tritones
127
pression by means of a single pair of pitch classes, say x and y, span-
ning clen c/2 as the interval from x to y and the interval from y to x.
The familiar case is, of course, the tritone of M12,7,where the spec-
trums of the traditional4th and 5th intersect, both including an inter-
val of six semitones. We borrow the traditionalterm and call any half-
"octave" interval a "tritone," regardless of its dlen.
Definition 2.4. A tritone is a two-element subset {Di, Dj} of DC,d
C
such that clen(Di, Dj) = clen(Dj, Di) = 2. Tritone is defined only for c
even.
Note that tritones are not necessarily ambiguous. A tritone
{Di, Dj}is unambiguousif dlen(Di, Dj) = dlen(Dj, Di) and ambiguous
if dlen(Di, Dj) $ dlen(Dj, Di).
Consider the sets D12,4 and M12,4pictured in fig. 7. Both have the
same chromatic and diatonic cardinalities, but their tritone contents
are very different. In one (fig. 7a) < 1 > n < 3 > = {6}, and in the
other (fig. 7b) < 2 > = {6};the former has one ambiguoustritone and
the latter, two unambiguoustritones. So we see that for pcsets in gen-
eral, knowing c and d does not tell us how many tritones the set has
or whether they are unambiguous or ambiguous. If we restrict our-
selves to ME sets, however, we can compute this information from c
and d, as the following lemmas show.
Lemma 2.6. If both c and d are even, then Mc,d has precisely d/2
tritones, all of which are unambiguous, and < d/2 > = {c/2}.
Proof. Since there are d elements in Mc,d, there can be at most
d/2 distinct tritones. Now consider = d/2: cl = c(d/2) = (c/2)d +
0. Thus all the intervalsof dlen d/2 have clen [ (d2) (theorem 1.11);
]
[c(d/2)
that is, there are d intervals of clen [c(d2)] = c/2. Thus there must
be at least d/2 distinct tritones. It follows that there are exactly d/2
tritones, and < d/2 > = {c/2}.
Further, since < d/2 > = {c/2} is a single-element spectrum, clen
c/2 cannot be ambiguous (lemma 2.5). .
Lemma 2.7. If Mc,d has a tritone ambiguity (implying c is even),
d-1 d+l
then d is odd, < <-- < > = {c/2}, and c/2 belongs to no
other spectrums.
Proof. Assume Mc,d has an ambiguity. If d is even, then there are
no ambiguous tritones (lemma 2.6). Thus d is odd. Since Mc,d has an
ambiguity,d > c/2 (lemma 2.2). Consider I = - .
= d -
c()_ 2 ) +2
128
(a) 10,6,8,10} (b) 10.3,6,91
<1> n <3> = {6}: <2> = 1(6:
one ambiguoustritone two unambigLuoustritones
-
Since d> c/2 >0, the element c(d )/2 + 1 is a member of
d-1 L
< -- > (theorem 1.11). But
c(d- 1)/2+ = c2 + [- + 1
d 21 +
129
elements on one side of {Di, Dj} must differ from the number of ele-
ments on the other. Thus dlen (Di, Dj) # dlen (Dj, Di) and hence
{Di, Dj} is ambiguous.
C
(i) Assume d - 2 and Mc,dhas a tritone. Then, as shown above,
the tritone must be ambiguous. But if there is an ambiguity then
C C
< d (lemma 2.2), contradictingthe assumptionthat d -< 2 It follows
that there are no tritones.
(ii) As shown above, any tritones must be ambiguous. It remainsto
show how many tritones there are when C< d < c. We know from
lemma 2.7 that if there are ambiguous tritones, then
( d-1c d (( + d\c
?i C 1)=
"\z -
and
[c(d- 1)/2] c
130
Definition2.5. Suppose c and d are odd and
d-1 d+1
< 2- > n < --2- > 0.
Then we say that Dc,dcontains a pseudotritone.
Note that pseudotritones are, by definition, ambiguous. Also, un-
like tritones, they are not associated, by definition, with a particluar
clen. In fact, a pseudotritone may correspond to more than one clen
in the same set, as in the next example.
Example 2.3. The two sets pictured in fig. 8 have the same chro-
d-1
matic and diatonic cardinalities. In both cases, -- = 2 and
d+l
--- = 3. Now in fig. 8a, < 2 > < 3 > = {3, 6}, indicating the
presence of pseudotritones. On the other hand in fig. 8b,
<2> < 3 > = 0, indicating there are no pseudotritones. Note
that fig. 8a is not a ME set, and fig. 8b is a ME set. The following
lemma shows that these sets exemplify the general situation: pseudo-
tritones occur only in non-ME sets.
Lemma 2.10. No ME set has a pseudotritone.
Proof. If c = d, then (c,d) = d and there are no ambiguities
(lemma 2.3) and hence no pseudotritones. Assume d < c. The largest
element in < > is[( d )/2 + 1. If a pseudotritoneexists, then
d1)/2 =
[c(d d 1+ [c(d+ 1)/2] (lemma 2.4).
d
But
Further,
[c(d+1)/2 c+ 1
d + c-d1
j-2 2dj
c+l c-d
>- + 0 (since -T > 0)
c+l
2 ?
131
(a) D9.5= 10,1,2,3,5 (b) M .5 = 1()01.3.5.7)
<2> = {2,3,5,6} <2> = {3.4}
<3> = 3,4,6,71 <3> = (5.6}
<2> n <3> = (3,6) <2> n <3> = null set
(pseudotritones) (nopseudotritones)
Figure 8
132
5. If c is odd and d even, then neither tritone nor pseudotritone is
defined for Mc,d-
It is clear from the above that if a ME set contains one or more
tritones, then either all of them are ambiguous or all are unambigu-
ous.
As we demonstrate in the next section, the case of Mc,d with c is
even, d odd, and c/2 < d < c (e.g., the usual diatonic set; see case 3
above) is of special interest. We need to prove one more lemma re-
garding ambiguities for this case.
Lemma 2.11. Mc,d has an odd number of ambiguous dens if and
c
only if c is even, d is odd, and < d < c.
Proof.
C
(=>) If Mc,dhas an ambiguity, then 2 < d < c (lemmas 2.2 and 2.3).
Further, <> n <I + 1 > 0 if and only if <d - I> n<d -
(I + 1) > =#0 (lemma 1.16), implyingM,,dhas an odd numberof am-
biguous dens if and only if for one of these dens, < I > = < d -
(I + 1)> (or equivalently <I +1> = <d-I>) and
<I>n<I + 1>#Q0.Butif<I> = <d - (I + 1)>then =
d-1
d - (I + 1), implyingI = 2-. Hence d must be odd. Further, if c is
d-1 d+l
odd, then < 2 > < > = 0 (lemma 2.10), implying by
lemma 1.16 an even number of ambiguousdens (contradiction).Thus
c must be even.
c
(<) If c is even, d is odd, and 2 < d < c, then there is an ambiguous
tritone (lemma 2.8), and < --- > n < - > -= (lemma 2.8,
d-1 d+l
- (
ii). It follows that < d > n < d d >2
= is also an ambiguous tritone. But < d - > n < d -
d+l- d+l d-1
> = < 2 > n< >, so these two pairsof intervalspec-
trums are one and the same. Thus if c is even, d is odd, and Mc,d has
an ambiguity, then it has an ambiguous tritone (lemma 2.9) and all
other ambiguitiesoccur in pairs (lemma 1.16). It follows that Mc,dhas
an odd number of ambiguous clens. o
Diatonic sets
133
Lemma 2.12. Let c and d be integers such that 1 < d < c. Then
there are precisely (c,d)values of I, 0 < I < d - 1, such that d divides
cd.
Remark. In number theory, a is said to "divide" b if and only if a
and b are integers, a - 0, and there is an integer x such that b =
ax. We write "a Ib" when a divides b, and "a - b" otherwise.
Lemma2.13. If(a, c) = 1 and (b,c) = 1, then (ab, c) = 1.
We now state some propertiesof spec(Mc,d). The next lemma shows
that the cardinalityof spec(Mc,d) is a function of c and d.
Lemma 2.14. I spec(Mc,d) I = 2d - (c,d).
Proof. From theorem 1.11 we know that < I > is a single element
spectrum if and only if d I cI; else < I> is a two-element spectrum.
Thus
I spec(Mc,d I = (number of single element spectrums)
+ 2. (number of two-element spectrums)
= (number of values of I such that d Icl)
+ 2 *(numberof values of I such that d (r) cI), 0 < I < d - 1.
But the number of values of I such that d IcI is (c,d) (lemma 2.12).
There are d possible values of I. Thus the number of values of I such
that d (-t) cl is d - (c,d). Therefore
Ispec(M,) I = (c,d) + 2(d-(c,d))
= 2d - (c,d). ?
Lemmas 2.15 and 2.16, given without proof, follow directly from
lemma 2.4 and the definitions of ME set, ambiguity, and spec(Dc,d).
Note that, by lemma 2.4, the elements of spec(Mc,d) have multiplicity
one or two.
Lemma 2.15. There are n ambiguous dens in Mc,d if and only if
there are n elements of multiplicity2 in spec(Mc,d).
Lemma 2.16. If I spec(Mc,d) I > c, then Mc,dhas at least one ambig-
uous clen.
Lemma2.17. If Ispec(Mcd) I > c + 1, then Ccis properlyincluded
in spec(Mc,d).
Proof. Recall that C, is the multiset {0, 1,2, . . .c-1}.
If Ispec(Mc,d) I - c + 1, then clearly Mc,dhas at least one ambiguity.
c
Thus < d (lemma 2.2), implying Cc is included in spec(M,d) (theo-
rem 1.12). Since Ispec(Mc,d)I > I Cc I = c, Cc is properly included in
spec(Mc,d).
The usual diatonic set has the following important property: each
of the possible dens {0, 1, 2, . . ,c - 1} appears exactly once in its
spectrum, except for clen 6, which is represented twice. The dual ap-
pearance of a clen in the spectrum signals the presence of an ambi-
guity. Thus spec(M12,7)is the smallest spectrum including all dens
134
while containing at least one ambiguity. The following definition for-
malizes this property, which is similar, though not identical, to Ag-
mon's "efficiency" condition.
Definition2.6. M(c) (read "min c") is the set of ME sets whose
spectrums are the smallest sets that properly contains C. That is,
Mc,d e M(c) if and only if both of the following hold true:
(1) Cc is properly included in spec(M,,t).
(2) For all M ,d such that CC is properly included in spec
(MCd), I spec(M,^() I I spec(Md) I.
We write Mc to indicate an element in M(c).
Example2.3. M(12) = M(12,7).
This example tells us that M(12)consists of precisely all the major-
scale sets.
Note that for a particularchoice of c, M(c) is not necessarily re-
stricted to a unique d. For example, Ispec(M1511)I
Ispec(M15.12)I = 21, and M(15) = M(15,11)U M(15,12)since the set
spectrums of M5.11 and M15.12are the smallest multisets which prop-
erly contain C15. If c = 3 (mod 6), the situation with M(c)is complex.
For example, M(3) = 0, M(9) = M(9,7), M(15) = M(15,11) U
M(15,12), M(21) = M(21,13). We conjecture that when c= 3 (mod
6) there is no finite set of algorithmsthat describe M(c). More specif-
ically, we conjecture that if 3 Ic and 5(+)c, then M(c) = M(c, (c +
5)/2); if 3 I c, 5 Ic, 7 ({) c, and 9(t) c, then M(c) = M(c, (c + 9)/2);
if 31 c c, 7 (Y) c, and 9 c, then M(c) = M(c, (c + 7)/2) U
M(c, (c + 9)/2), etc. It would appear that M(c) depends on how c fac-
tors into powers of odd primes.
Fortunately, if c is even or if c-1,5 (mod 6), then M(c) is re-
stricted to a unique d, as we show in theorem 2.1 below. In prepara-
tioin for this theorem, we state the following lemma without proof.
Lemma 2.18. Let a, b, c, x, and y be integers such that ax +
by = c. Then (x, y) divides c.
Theorem 2.1. Let c > 4.
(i) If c is not congruent to 0 (mod 4), then M(c) = M(c, c/2 +
1) and I spec(M) I = c + 1.
(ii) If c is not congruent to 2 (mod 4), then M(c) = M(c, c/2 +
2) and Ispec(M) I= c + 3.
(iii) If c is not congruent to 1, 5 (mod 6) (or equivalently if c is odd
and c is not congruent to 0 (mod 3)), then M(c) = M(c, (c +
3)/2) and Ispec(M,) I = c + 2.
Proof. In general, to prove M(c) = M(c,d) for a particular value of
d, we first compute I spec(M,d) I. We then show that there is no d*
such that c < I spec(M ,d!) I < I spec(M,,d) I. This implies M(c,d) is in-
cluded in M(c). Finally, we show that if Mc,.* E M(c), then
Mc,d E M(c,d).
135
(i) Assume c- 0 (mod 4). Then (c/2, c/2 + 1)
1 and (2, c2 + 1) = 1, implying (c, c/2 + 1) = 1 (lemma 2.13). Thus
Ispec(Mc,/2+l) = 2(c/2+ 1)-1 (lemma 2.14)
=c+ 1.
Since Cc is included in spec(Mc,12+1) (lemma 2.17) and
Ispec(M,c/2+) I = c + 1, we conclude that Mc,c2+1E M(c). Thus
M(c, c/2 + 1) is included in M(c). Further, if M d E M(c), then
I spec(M,d) I = c + 1. Thus 2d - (c,d) = c + 1 (lemma2.14). But
then (c,d) divides d, (c,d), and c; hence (c,d) must divide 1, implying
(c,d) = 1. It follows that
2d-1 = c + 1
d = c/2 + 1.
Hence M,d M(c, c/2 + 1), implying M(c) is included in
M(c, c/2 + 1). We conclude that M(c) = M(c, c/2 + 1).
(ii) Assume c = 2 (mod 4). Then (c/2 + 2) - c/2 = 2, implying
(c/2, c/2 + 2)divides 2 (lemma 2.18). Thus (c/2, c/2 + 2) = 1,2. But
c/2 is odd, implying(c/2, c/2 + 2) = 1. Further,c/2 + 2 is odd, imply-
ing (2, c/2 + 2) = 1. It follows that (c, c/2 + 2) = 1 (lemma 2.13).
Now we compute Ispec(Mc,c2+2)I.
Ispec(Mc,c2+2)1 = 2(c/2 + 2)- 1 (lemma 2.14)
=c + 3.
136
(c,d). But then 4 must divide c, implying c 0 (mod 4) (contradic-
tion). It follows that M(c, c/2 + 2) is included in M(c).
LetMc,d EM(c). Then I spec(Mc,d) I = c + 3.Thus
2d-(c,d) = c + 3.
Since(c,d) divides d, (c,d), and c, it must divide 3. It follows that
(c,d) = 1,3. If (c,d) = 3,then
2d-3 = c+ 3
d = c/2 + 3.
But since c 2 (mod 4), c/2 + 3 must be even. Thus 2 divides both c
and d, contradictingthe assumptionthat (c,d) = 3. Hence (c,d) = 1.
But then
2d-1 =c+ 3
d = c/2 + 2.
Thus Mc,dEM(c, c/2 + 2), implying M(c) is included in M(c, c/2 +
2). It follows that M(c) = M(c,c/2 + 2).
(iii) Assume c- 1,5 (mod 6). Then 2((c + 3)/2) - c = 3. Thus
(c, (c + 3)/2) divides 3 (lemma 2.18), implying (c, (c + 3)/2) = 1,3.
But if c is not congruent to 0 (mod 3), 3 does not divide c. Thus
(c, (c + 3)/2) = 1. Then
I spec(Mc,c+3)/2)I = 2((c + 3)/2)- (c, (c + 3)/2)
=c + 3-1
=c + 2.
Since c is odd, Mc,dmust have either no ambiguous dens or an even
number of ambiguous dens (lemma 2.11). Thus we cannot have
Ispec(Mc,d) I = c + 1 (lemma 2.15). We conclude that Mc,(c + 3)/2
is included in M(c).
Now assume Mc,d M(c). Then
I spec(Mc,d) I = c + 2
2d- (c,d) = c + 2 (lemma 2.14).
Since (c,d) divides d, (c,d), and c, (c,d) = 1,2. If (c,d) = 2, then c
must be even (contradiction). Hence (c,d) = 1. Then
2d-1 = c + 2
d = (c+3)/2.
Thus Mc,d M(c, (c + 3)/2), implying M(c) is included in M(c, (c +
3)/2). It follows that M(c) = M(c, (c + 3)/2). ?
137
The case c 0 (mod 4), covered in part 1 of theorem 2.1 above, is
of particularimportance as we shall see.
The presence of exactly two half steps (intervals of clen 1) is an
important characteristic of the usual diatonic set. The next lemma
shows that ME sets with two half steps occur only under a particular
condition.
Lemma 2.19. Mc,d has precisely two intervals of clen 1 if and only
ifc = 2(d - 1).
Proof. It is clear that clen 1 can only be associated with dlen 1, for
if clen 1 is associated with, for example, dlen 2, then
2c
7<2
c < d (contradiction).
If
(=) Mc,d has an interval of clen 1, then 1 < 1>, implying
[ ] = 1.Thus
a<2
c<2d.
Hence, by theorem 1.11,
c. =d+(d - 2)
c = 2(d-1).
(<) If c = 2(d - 1),then
c-1 = d + (d-2),
implying Mc,d has precisely two intervals associated with dlen 1 that
have clen 1. Since clen 1 can only be associated with dlen 1, if follows
that there are precisely two intervals of clen 1.o
Theorem 2.2 below characterizesthe class of ME sets with exactly
one ambiguity by stating ten equivalent conditions. Conditions 1-4
deal with ambiguity; conditions 5-7 are different ways of stating the
relationshipbetween chromaticand diatonic cardinalities;condition 8
deals with "half steps"; conditions 9 and 10 deal with the spectrumof
the set.
Theorem 2.2. For Mc d the following are equivalent:
(1) Mc,d has precisely one ambiguity.
(2) Mc,d has precisely one ambiguous clen.
(3) Mc,d has precisely one ambiguous tritone.
138
(4) Mc,dhas precisely one tritone and d - 2.
(5) c = 2(d- 1) and (c,d) = 1.
(6) c = 2(d - 1) and disodd.
(7) c = 2(d - 1) and c0 (mod 4).
(8) Mc, has precisely two intervals of clen 1, and c 0
(mod 4).
(9) M(c,d) = M(c), and c 0 (mod 4).
(10) I spec(Mc,d) I = c + 1
But then cois a member of < Io > intersect < Io + 1 >, implyinga sec-
ond ambiguous clen. Thus if there is precisely one ambiguous clen,
then there is precisely one ambiguous tritone.
(3 > 4) Assume there is precisely one ambiguous tritone. Then d
must be odd (lemma 2.7). Hence d - 2 and M,,d can contain no un-
ambiguoustritones (lemma 2.8). It follows that M,Ccontains precisely
one tritone.
(4 > 5) Assume Mc,d has precisely one tritone and d- 2. If d is
even then Mc,d has d/2 tritones (lemma 2.6). Since Mc,d has precisely
one tritone, d/2 = 1, implying d = 2, contradictingthe assumption.
139
Thus d must be odd; hence (2, d) = 1. Further,since there is precisely
C
one tritone, d - = 1 (lemma 2.8). It follows that (c2, d = 1 and
2
c = 2(d- 1). Since (2, d) =1, (2(Q), d) = (c,d) = (lemma 2.13).
(5 = 6) Assume (c, d) = 1 an c = 2(d - 1). Then d must
be odd, else (c,d)- 2.
(6 => 7) Assume d is odd and c = 2(d - 1). Then d - 1 is even
and hence 4 12(d - 1). It follows that c-= 0 (mod 4).
c
(7 9) Assume c = 2(d - 1) and c 0 (mod 4). Then d = 2 +
c
1, implying M(c,d) = M(c, + 1) = M(c) (theorem 2.1) and
2
c O (mod 4).
(9 = 10) Assume M(c,d) = M(c) and c 0 (mod 4). Then Ispec
(Mc) I = c + 1 (theorem 2.1). Thus I spec(M,d) I = c + 1.
(10 d> 1) Assume Ispec(Mc.d) I = c + 1. Then C, is properly in-
cluded in spec(M, d) (lemma 2.17). Since I CcI = c, and every element
of C, has multiplicity 1, there must be precisely one element of
spec(Mc,d) that has multiplicity 2. Thus there must be precisely one
ambiguous clen (lemma 2.15). It follows that c is even and d is odd
(lemma 2.11), implying the ambiguous clen is a tritone (lemma 2.9).
We must now determine how many tritone ambiguities there are.
Since I spec(M,d) I = c + 1, we know that 2d - (c,d) = c + 1
(lemma 2.14). Further, since (c,d) divides c and d, (c,d) must divide 1.
Hence (c,d) = 1. It follows that
II +
~~~/
/x_Dh / 2= 0+1
Figure 9
140
2d-1 = c + 1
d-c/2 = 1.
141
(a) tritone (b) attachhalfstepsto tritone
Lemma 2.20. Let t be a tritone {Di, Dj}. There are exactly two
(transpositionallyequivalent) diatonic sets which include t.
Proof. By definition and by theorem 2.2, a diatonic set has exactly
d-1
one tritone. By lemma 2.7 either dlen(Di, D) = - and
d+l
dlen(Dj, Di) = -Z or these assignmentsare reversed. The two pos-
sibilities imply two distinct (transpositionally equivalent) diatonic
sets. *
142
(a) tritone (b) attachhalf steps to tritone
143
MA, this would imply that MA has more than one tritone, contradict-
ing the assumption that it is a diatonic set. Similarly, for any tritone
except t, if one of the pcs of the tritone is in MA, the other is not. Now
since MA and MBarediatonicsets, I MA I = MB I = c/2 + 1 = 2 +
(c/2 - 1). Therefore MA and MB each contain the two pcs of t plus
exactly one pc from each of the (c/2 - 1) other tritones. It follows that
D is in either MA or MB but not both. ?
Generated ME sets
C,d
144
Examples 3.2. The generators of the usual diatonic set are 5 and
12-5 = 7. The generators of the usual whole-tone set are 2 and
12-2 = 10.
Note that any set of diatonic cardinality0, 1, c - 1, or c, is max-
imally even. In this sense we can regard sets of these cardinalitiesas
triviallyME and other ME sets as nontriviallyso. This distinctionwill
help to simplify the following treatment of generators.
Lemma 3.2. If g is a generator of Mc,d (nontrivial), then g is an un-
ambiguous clen.
Proof. Assume g is an ambiguous clen. Then
Further,
and
g = + 1
t
and d - r2 intervals associated with I + 1, whose clen is
But then
cI = d(g-1) + rl
rl = cl-dg + d
and
c(l + 1) = dg + r2
d-r2 = dg + d-cl-c.
Since g is a generator, at least d - 1 intervalsmust have clen g. Hence
rl + (d-r2)-d-1
(dg + d-cl-c) + (cl- dg + d)- d-1
d c- 1,
145
contradictingthe hypothesis that Mc,d is nontrivial. It follows that g is
an unambiguousclen. ?
Lemma 3.3. Suppose (c,d) = 1. If Mc,d has a generator, then
(c,d) = d.
Proof. If g generates M,d, then all intervals with clen g are asso-
ciated with the same dlen, say I (lemma 3.2). If I has a two-element
spectrum, then there is one interval associated with I that has clen
g? l.So
cl = gd + 1
or
cl = gd + (d-1) (theorem 1.11).
In either case, since (c,d) divides c and d, (c,d) must also divide 1,
contradictingthe hypthesis that (c,d) - 1. It follows that all intervals
associated with I have clen g. Therefore
dg = cl (lemma 1.2).
Now consider two ways of representing Mgd:
Dk,+
-
Dk, [c/d] + 1 (mod c)
Dk,+1 - Dk2 [cld] (mod c).
But since Md, is a generated ME set, there exist integers mn1,n, m2,
and n2 such that
Dk,+l j + mg (mod c)
Dk =j + n1g (mod c)
Dk2+l j + m2g (mod c)
Dk j + n2g (mod c).
146
Thus
Dk1+ - Dk (m, - n,)g- [cld] +1 (mod c)
Dk2+l
-
Dk2 (m2 - n2)g
[cld] (mod c).
Subtractingthe second congruence from the first we get
(ml - nl - m2 + n2)g - 1 (mod c).
It follows that
(g,c)= 1.
147
Example3.3. M721 is generated by clen 7. However, there are not
d - 1 = 10 intervals of clen 7 associated with one particular dlen.
Hence clen 7 must be ambiguous. In fact, 7 E <6>n<7>.
The following lemma and corollary, given without proof, are well-
known in number theory.
Lemma 3.5. If (c,d) = 1, then there exist unique integers g1 and g2,
1 < g1, g2 < c - 1, such that dg1 - 1 and dg2 = -1 (mod c).
Corollary 3.1. Let c, d, gl, g2 be as in the above lemma. Then
g2 = c - gl.
As established in part 1, there are three classes of ME sets based on
(c,d): class A, (c,d) = 1; class B, (c,d) = d; and class C, l<(c,d)<d.
The following theorem shows that the generator of M, d if there is
one, is dependent upon (c,d).
Theorem 3.1. Let MC,d = {Do, D ..., DdZ_}. Then
A. The following are equivalent.
(1A) (c,d) = 1.
(2A) All nonzero interval spectrums have two elements.
(3A) gl, g2, as defined in lemma 3.5 exist and are generators.
B. The following are equivalent.
(1B) (c,d) = d.
(2B) All interval spectrums have one element.
(3B) cld and c-cld are generators.
C. The following are equivalent.
(1C) l<(c,d)<d.
(2C) There is at least one nonzero spectrumwith one element
and one with two elements.
(3C) Mc,d has no generator.
Proof.
A. (1A=>2A)Assume (c,d) = 1. Then every nonzero spectrumhas
two elements (lemma 1.3).
(2A:=3A) Note that <dl(c,d)> = {cl(c,d)} (lemma 1.7). Since
every nonzero spectrum has two elements, we must conclude that
(c, d)= 1, else there would be a nonzero one-element spectrum,
namely <dl(c, d)> (lemma 1.7). Hence gl and g2 as in lemma 3.2
exist. Consider gI with dg =- -1 (mod c). There exists an I, 1 < I <
d - 1, such that cl = dg, + 1. It follows that, for the set of d intervals
associated with I, there are precisely d - 1 of them with clen g, (the-
orem 1.11). Thus gl is a generator (lemma 3.4). By lemma 3.1 and
corollary 3.1, MC,d is also generated by g2.
(3A-1A) Clearly if dgl -1 (mod c) or dg2 1 (mod c), then
(c, d)= 1.
B. (1B:>2B) Assume (c,d) = d. Then <I> = {cIld} (lemma 1.7),
implying every spectrum has one element.
148
(2B=3B) Assume every spectrum has one element. Consider the
intervals of dlen 1. By theorem 1.11, c * 1 = dq. Thus dI c and
<1> = {c/d} (lemma 1.7). It follows that
MC,d = {Do, Do + cld, Do + 2(c/d), .. , Do + (d - l)(c/d)}
and hence cld is a generator. Further, c - cld is a generator (lemma
3.1).
(3B=>1B)If cld and c - cld are generators, then cld must be an
integer. Thus (c,d) = d.
C. Here the method is to show that 1C and 2C imply each other,
and that 1C and 3C imply each other.
(1C=2C) Assume 1 <(c,d)<d. Then d does not divide c, and hence
c * 1 = dq + r, where 1 - r - d - 1. It follows that <1> is a two-
d
element spectrum (theorem 1.11). If I = c-,d) then <I> is a one-
element spectrum (lemma 1.7). Thus M,. has both one- and two-
element nonzero spectrums.
(2C=>1C)Assume that there is at least one 1-element nonzero spec-
trum and at least one 2-element spectrum. If (c,d) = 1, then there are
no 1-element nonzero spectrums(part A); and if (c,d) = d, then there
are no 2-element spectrums (part B). Hence 1 < (c,d) < d.
(1C=3C) This follows immediately from lemma 3.3.
(3C=1C) From parts A and B we know that if (c,d) = 1 or (c,d) =
d, then Mcd is a generated set. Hence if Mc.d is not a generated set,
then 1 < (c,d) < d. ?
Theorem 3. 1 will be important in relation to complementaryME
sets, considered in the next section. We will also need to consider triv-
ial ME sets.
Theorem 3.2. Trivial ME sets are generated.
Proof. Since neither Mco nor MC,,has a second element which can
be generated with reference to a first element, we may consider any g,
O - g < c - 1, to be a generator. Any clen that is coprime to c is a
generator for Mc.c_- or M,,.
ME complements
149
We will usually speak, not of literal complementation (as in the
case of white and black keys), but of "abstract"complementation. As
in atonal set theory, if each of two set classes includes one of a pair of
literally complementarysets, then we say that the two set classes are
abstractlycomplementary.For example, M(12,7) and M(12,5) are ab-
stractlycomplementary, since they contain, respectively, the C-major
set {0, 2, 4, 5, 7, 9, 11} and the literally complementarypentatonic set
{1, 3, 6, 8, 10}. Where the terms literaland abstractare omitted in our
discussion, the distinction will be clear from context.
Definition 3.2. The complement of D,d, compl(D.d), is Uc\D,d
(the set consisting of all elements of Uc not in D,, ). Note that the
cardinality of compl(Dc,d) is c - d.
We will state without proof two lemma easily proved in elementary
number theory, and then prove that the complement of a ME set is
itself a ME set.
Lemma 3.6. For any integers p and d there exist integers q and r,
0 c- r < d
c-7 - 1, such that
c = dq + (c,d)r.
Lemma 3.7. Let c, d, x, y, z be integers such that
x y z
c+d =(c,d)
Then c divides (c,d)x and d divides (c,d)y.
Theorem 3.3. Jcd and JCl-dl, 0 < m < c - 1, are complementary
sets; that is, JCd = compl(J d).
Proof. Let m = j(c,d) + r, where 0 < r < (c,d) - 1. Then
c- m- =(c- (j+ 1)(c,d)) + ((c,d) - r - 1)
(where 0 < (c,d) - r - 1 - (c,d) - 1).
= = J c-(j+(cdl
So Jm,d jlc and Jc-dl (lemma 1.11). Thus we need
only show that JCCdd)and Jc (+ l)cc,d)are complementary sets. But since
1 (c,d)
I c,d = d andcIJc = c - d, these two sets are literally com-
plementary if and only if they have no elements in common.
Assume, to the contrary, that the two sets have one or more ele-
ments in common. By definition the elements of Jfc,d) and
-lj+)(c,d)
[c
+ (c , 0 -
cM +cd - are )c,d)] N d 1, and
cM +
c -d(,+ )(c,d 0 -< M - c - d- 1, respectively. Then
by assumption there exist M and N such that
cN + j(c,d) [cM + c-( + l)(c,d)
d c-d PI
150
for some pitch class p.
Then by lemma 3.6
cN + j(c,d) = dp + (c,d)rl,
-
where 0 < rl d/(c,d) - 1, and
cM + c-(j + l)(c,d) = (c-d)p + (c,d)r2,
where 0 - r2 - d/(c,d) - 1) - 1, noting here that (c, c - d) = (c,d).
Now, adding the remainder inequalities, we get
C
0 -- rl + r2 --(c-- - 2.
(c,d)
With some manipulation we solve cN+j(c,d) = dp + (c,d)rI for p.
Substitutingappropriatelyin cM + c-(j + 1)(c,d)=(c-d)p + (c,d)r2
we obtain
r2 + r2 + 1 (c/(c,d))N + j + r _ M + N + 1
c + d (c,d)
Thus c divides (c,d)(rl + r2 + l)(lemma 3.7). Clearly (c,d)(rl+
r2 + 1) = 0; therefore
(c,d)(rl + r2 + 1)-c
and
C
rl + r2 '- c,-1,
151
(b) If (c,d) 1 and 0 < d < c/2, then c - > c - d > c -c/2, im-
plying c/2 < c - d < c. Thus 1 < (c, c- d)<c - d (since c,d) = (c,
c - d) and d < c/2), implying compl(Mcd) = Mcc _ d has no generator
(theorem 3.1, C). Thus if Mc d and compl(M,cd) are both generated
ME sets, then either (c,d) = 1 or d = 0, c/2, c.
(~)
(a) If (c,d) = 1, then (c,c - d) = 1, and both Mc d and compl(Mc,d)
are generated ME sets (theorems 3.1, A and 3.2).
(b) If d = 0, or d = c, then the corresponding ME sets and their
complements are trivially ME, hence generated (theorem 3.2).
(c) If d = c/2, then (c, c-d) = (c,c12) = c/2, and both Mc,/2 and
compl(Mc.c/2) are generated ME sets (theorem 3.1, B). ?
Having gained some understandingof ME set complements, we are
now at a logical place to recall the constructionof ME sets by means
of points distributedaround the circumferenceof a circle. The follow-
ing theorem proves the claim that this construction,firstpresented fol-
lowing example 1.4, yields a ME set and its complement.
Theorem3.5. Let c and d be positive integers such that d < c. Place
d white points equidistantly around the circumference of a circle.
Then place c - d black points equidistantlyaround the circumference
so that no two points (one white, one black) are in the same location.
Assume that intervals from one point to another are measured clock-
wise aroundthe circumference.Assign clen 1 to every adjacent pair of
points, regardlessof color. Then both the set of white points and the
set of black points represent ME sets in a chromatic universe of car-
dinality c.
Proof. Clearly if d = c - d, then the theorem holds true. If d = c
- d, we may assume without loss of generality that d > c - d. The
general approach to the proof will be as follows:
(a) Find the smallest clockwise distance from a white to a black
point; call this distance 8 (fig. 12a).
(b) Call the white point a0. Moving clockwise around the circle, call
the other points a1, a2, . . ., a , consecutively. Assume that the dis-
tance aroundthe circumferenceis d. Calculatethe distancesfrom a0 to
all points in terms of 8 and d (fig. 12b), and label each point with its
respective distance from ao.
(c) Define an order-preservingmap u that sends aj to j (fig. 12c).
(d) Show that o maps the black points (and hence the white points)
to a ME set.
After points have been labeled with their distances from a,, the sets
152
(a) c= 9, d =5
0 5 O (a o)
?
Oo (a() (a))
14 8 a-) a (7) 2
5 4
153
rjd jd
c- d] < + - c <[ -d
jd1
d+1,
where 0 < j < c - d - 1. Let Sc = Sc,d U Sc,c-d Then
[c -d +j+ 1.
We now show that ('(Scd) and C(S,,c-d) are ME sets.
d 6 s
6=" (c - d- 1)d
O'(Sc,cd) = 0(8), (85+ c - d) . . . (c d
{[0- d] + 1, [ d]
- d) -
+ 2, .... [ + (c+d)}
c + (c - d)] 1 c + (c- d)
lL
c- d [[ c-d d
[(c-d-1) c + (c-d)]
c-d
c-d-d
={Jc 0, JJ,(1)
=JC,c- d(?) c,c....,
-d( -d JCCC -c d d - 1})
154
spectrumsare the largest proper subsets of Cc. That is, Mc,d E M(c) if
and only if both of the following hold true:
(1) Spec(M, d) is properly included in Cc
(2) For all M,d such that spec(M,d) is properly included in CC,
I spec(Mc,d) I I spec(Mcd) I.
We write Mc to indicate an element in M(c).
Example 3.5. M(12) = M(12,5), the usual pentatonic set.
Spec(M12,5)= {0, 2, 3, 4, 5, 7, 8, 9, 10} = C12\{1, 6, 11}.
The question naturallyarises: for any particularchoice of c, is there
a unique d such that M(c,d- = M(c)?In contrastto the situation with
M(c), examined previously, there are no problem cases here, as the
following theorem shows.
Theorem 3.6. Let c 2.
(i) If c = 2, 4, 6, then M(c) = M(c,c/2) and I spec( Mc)I = c/2.
(ii) If c is odd, then M(c) = M(c,(c - 1)/2) and
I spec(Mc) I = c - 2.
(iii) If c is even and c- 8, then M(c) = M(c,c/2 - 1) and
(a) if c 0 (mod 4), then I spec(Mc)I= c - 3, or
(b) if c 2 (mod 4), then I spec(M,)l = c - 4.
Proof. In general, to prove M(c) = M(c,d) for a particularvalue of
d, we first compute I spec(Mcd) I. We then show that there is no d*
such that d* - d and I spec(M, d) I - I spec(M, d*) I < c.
(i) We can show by computation that for c = 2, 4, 6, M(c) =
M(c,c/2). Further,
I spec(Mc c/2) I =2 (c/2) - (c, c/2)(lemma 2.14)
=c - c/2
= c/2.
155
Thus Mc,d (i~M(c). It follows that M(c) = M(c,(c - 1)/2).
(iii) Assume c is even and c :-8.
We can show by computation that M(8) = M(8,812 - 1) and
I spec(M8,3) I = 8 - 3. Thus for c = 8 we have M(c) = M(c,c12 -
1) and I spec(Mc, c/21) I= C- 3.
Now suppose c > 8. As in (i) above, d ? c12 (theorem 1.12) and
Mcd has no ambiguities (lemma 2.2). Hence spec(Mc,d) is included in
Cc. We now consider the two cases (a) c 0 (mod 4) and (b) c 2
(mod 4).
(a) If c 0 (mod 4), then d = c12 - 1 must be odd. Thus (2, d) = 1.
Further, d - c12 = -1, implying (c12, d) = 1 (lemma 2.18). Thus
(c, d) = 1 (lemma 2.13), and
I spec(M,12-1) I =2(c/2 - 1) - (c, c/2 - 1)
= (c - 2) - 1
c - 3.
(b) If c 2 (mod 4), then d = c12 - 1 must be even, implying (c,
d) : 2. Also c - 2(c12 - 1) = 2, implying (c,d) ? 2 (lemma 2.18).
Thus (c,d) = 2 and
I spec(Mh,,c/21)I = 2(c12 - 1) - (c,d)
=(c - 2) 2
c - 4.
In either case
I spec(M,,,/2 ) I c - 3.
Thus spec(M,, 2 -1) is properly included in C, (lemma 2.2). Further,
I spec(M,,c/2-1) I : c - 4.
Now we must show that for d = c12 - n, where 2 - n - c/2,
I spec(M,,,2-n) I< c - 4.
But
I spec(M,c/2,-) I = 2(c12 - n) - (c, c12 - n)
(c - 2n) - 1 (since (c, c/2 - n) ? 1)
c - 5 (since n ? 2)
<c -4.
Hence if d < c/2 - 1, then I spec(Mc,d) I < I spec(M,.2,,/2-) 1. Now we
consider d = c/2.
I spec (Mc,02) I = 2(c/2) - (c, c/2)
c - c/2
c12
<c - 4 (since c > 8).
156
Thus I spec(Mc,c2) I < I spec(Mc,,/2-) I. Since we can show by com-
putation that M(8) = M(8,3), we conclude that if c is even and c - 8,
then M(c) = M(c,c/2 - 1). ?
We will return to our investigation of the properties of M(c) and
M(c) later. First we explore some further relationships between dia-
tonic sets and their complements.
Recall that diatonic sets have precisely one tritone. Indeed we used
this property in the construction of diatonic sets. In the case of dia-
tonic set complements, the clen c/2 - 2 (the major 3d in the usual
diatonic set) plays a similarrole-it is the singularintervalof this clen.
Lemma 3.8 below shows that ME sets with precisely one interval of
this clen are restricted to particularrelationshipsbetween c and d. In
preparation for the lemma, note- r<
that up to this point we have been
writing cl = dq + r where 0 d, attachinglittle or no importance
to the quantity q; however
cI = dq + r= d[ ] + r.
Lemma 3.8. Let c > 8. Then Mcd has precisely one interval of clen
c/2 - 2 if and only if (c,d) = 1 and either c = 2(d + 1) or c = 2(3d -
1).
Proof.
(>) c - 8 implies c/2 - 2 - 2, which in turn implies that clen c/2
- 2 is in a nonzero interval spectrum (theorem 1.11); also note that
cl~
- > 0 implies I = O.) If Mc. has precisely one interval of clen c/2 -
, fien there exists an interval of dlen I such that c/2 - 2 E <I> and
cl = d[- + 1 or c = d[- + (d - 1) (theorem 1.11). In either
case since (c,d) divides c and d, (c,d) must divide 1; thus (c,d) = 1,
and clearly d # c/2, else (c,d) = 1. Thus either d > c/2 or d < c/2. We
consider each of these cases in turn.
Case 1. Suppose d > c/2. Then Mc, has a tritone ambiguity and
2\ 1> n d 2 / = {c
(lemma 2.8). Since (c,d) = 1 every non-
zero spectrum has two elements (lemma 1.3). It follows that c/2 - 1
E 22 1> Hence c/2 - 2 E <d 2 >, since both c/2 and c/2 - 1
- 3)/2)
are in <d-~ 2 >. Then either [c((d d c/2 - 2 or
- 3)/2)]
-c((d
(( d )] + 1 = c/2 - 2(theorem 1.11). We nowconsiderthese
5wo subcases in turn.
-
(i) If[c((d 3)/2 2, h
157
c((d - 3)/2 = d[((d 32 + d - 1 (theorem 1.11)
= d(c/2 - 2) + d - 1
3c = 2d + 2
-2 c + 2 (since d - c)
c - 2 (contradiction).
-
(ii) f [c((d 3)/2) + 1 = 2 - 2, then
c((d - 3)/2) c/2
and
-
c((d - 3)/2 = d[c((d + 1 (theorem 1.11)
= d(c/2 - 3) + 1
3c= 6d -2.
But this would imply 3 divides 2 (contradiction).
Case 2. If d < c/2, then there are no ambiguities (lemma 2.2) and
no tritones (lemma 2.8). Since (c,d) = 1 every nonzero spectrum has
-
two elements (lemma 1.3). Hence <d 2 1>has two elements and has
no tritone. Thus c/2 - 2 E <d -2 > It follows that either
[c((d - 1)/2) / c((d - 1)/2)1
[ d c/2-2 or d + 1 = c/2 - 2(theorem
1.11).We now consider these two subcases in turn.
-
(i) If[c((d 1)/2) 2 2, then
158
(<=) From (c,d) = 1 and either c = 2(d + 1) or c = 2(3d - 1), it
follows that c must be even and d odd. Further, every nonzero spec-
trum has two elements (lemma 1.3) and Mc,d has no tritones (lemma
2.8). Hence if c/2 - 2 is in any interval spectrum, it must be in
d - 1>
2
Case 1. Assume c = 2(d + 1). Then
-
[cl] [c((d1)/2)d
= c/2 - 2
+ c- 2 (since c = 2(d + 1))
159
If a set contains precisely one interval of clen c/2 - 2, it must con-
tain precisely one interval of clen c - (c/2 - 2) = c/2 + 2, and the
converse is also true (theorem 1.11). Therefore, in the above lemma
and henceforth, whatever is implied by a singularinstance of c/2 - 2
is also implied by a singular instance of c/2 + 2.
Examples3.6. The case c = 2(d + 1)from lemma 3.8 is exemplified
by M(16,7); the case c = 2(3d - 1) is exemplifiedby M(16,3). Each of
the ME sets in these classes has exactly one interval of clen 6 = 16/2
- 2.
We are now in a position to prove the characterizationtheorem for
diatonic set complements. The equivalent conditions enumerated for
diatonic set complements are to some extent "complementary" to
those enumerated for diatonic sets in theorem 2.2; also, as in the case
of diatonic sets, the condition c = 0 (mod 4) is central here.
Theorem 3.7. For Mc, with c - 8, the following are equivalent:
(1) M, d is the complement of a diatonic set.
(2) Mc,d has precisely one interval of clen c/2 - 2 and c : 2(3d
- 1).
(3) c = 2(d + 1) and (c,d) = 1.
(4) c = 2(d + 1) and d is odd.
(5) c = 2(d + 1) and c 0 (mod 4).
(6) M, d has precisely two intervals of clen 3, and c 0 (mod 4).
(7) M(c,d) = M(c), and c 0 (mod 4).
C
(8) c is even and spec(Mc, ) = Cc {1, , c - 1}.
(9) I spec(MC,d) I = c - 3.
Proof. We will show 2 < 3, then 1 >3 = 4 > 5 6 > 7 => 8=
9 1.
(2 < 3) This follows directly from lemma 3.8.
(1 = 3) If Mc d is the complement of a diatonic set, then
compl(Mc,d) = Mc,c-d is a diatonic set. Then c = 2((c - d) - 1), and
(c, c - d) = 1 (theorem 2.2). Thus c = 2(d + 1) and (c,d) = 1.
(3 => 4) Assume c = 2(d + 1) and (c,d) = 1. Then d must be odd,
else (c,d) - 2.
(4 = 5) Assume c = 2(d + 1) and d is odd. Then d + 1 is even,
hence 4 divides 2(d + 1). Thus c= 0 (mod 4).
(5 = 6) Assume c = 2(d + 1) and c 0 (mod 4). It follows that
c = 2d + 2. But then
c1= [ ]+ 2
and there must be two intervals associated with dlen 1 whose clen is
[c- + 1 = 3(theorem 1.11). Since Mc, hasnoambiguities(lemma
160
2.2), there must be precisely two intervals of clen 3 and, by assump-
tion, c- 0 (mod 4).
(6 = 7) Assume Mcd has precisely two intervalsof clen 3, and c 0
(mod 4). If 3 is an ambiguous clen, then there exists an I, 0 < I d
- 1, suchthat <I> n<I + 1> = {3}and[j] + 1 = [c(d ] =3
(lemma 2.4).
But since 3 is associated with both I and I + 1, and since there are
precisely two intervals of clen 3, we have
I=[jd + d - 1)
(theorem 1.11)
c = 3d- 1
and
c(I + 1) = 3d + 1.
Putting these two equations together, we get c = 2, contradictingthe
assumption that c - 8. Hence 3 cannot be an ambiguous clen.
Since both intervals of clen 3 must be associated with the same
dlen, say I, then either [] = 3 or []
+ 1 = 3 (lemma 1.9). If
[] = 3,then
161
c 2 = 2d + 2
c=d+ 1.
Since c > 8, we have d - 7. Hence
c(i + 1) c 31
[ d ] [j-
= 3 + [a] (since c = d + 1)
=3 (since d 2 7).
But then 3 would be an ambiguous clen (lemma 1.9) (i.e., 3 E
<2>n<3>, contradiction). It follows that I = 1 and c ? 1 = 2d + 2.
Thus d = c/2 + 1, and since c 0 (mod 4), M(c,d) = M(c) (theorem
3.6).
(7 4 8) Assume c =0 (mod 4) and M(c,d) = M(c). Then d = 2 -
c
1 (theorem 3.6). Thus d < 2, implyingMCd has no ambiguities(lemma
d=
c c
But then (c,d) = 2, implying 2 = 3. Thus c = 6, contradictingthe as-
sumption c - 8. If (c,d) = 1, then
162
2d - (c,d) = c - 3
2d - 1= c - 3
d=2-1.
But then compl(M,,d) = Mc c-d, and
c - d = c - (c/2 - 1) (since d = c/2 - 1)
c = 2[(c - d) - 1].
Hence c is even, and since (c,d) = 1, (c, c - d) = 1, implying c - d
is odd. Thus Mcc-d is a diatonic set (theorem 2.2). It follows that Mcd
is the complement of a diatonic set. ?
In light of theorem 3.7, we now introduce some logical extensions
of our terminology for diatonic sets.
Definitions3.4. We say that the complement of a diatonic set is a
pentatonic set; the complement of a usual diatonic set is a usual pen-
tatonicset; the complement of a hyperdiatonicset is a hyperpentatonic
set.
Note that with the exception of the usual pentatonic ("black key")
sets, which have five notes, the infinitely many other pentatonic sets,
which we call hyperpentatonic, are not 5-note sets. Note also that, in
the 12-note chromatic universe, the class pentatonic does not include
5-note sets other than the usual pentatonic. While these possible con-
fusions are unfortunate, they are outweighed in the present context by
the symmetry between definitions for diatonic and pentatonic sets,
which reinforces our results on complementation. To avoid one of the
pitfalls noted above, we will use the cautionaryterm (hyper)pentatonic
instead of pentatonic.
The extremal sets M(c) and M(c) play "complementary"roles in
characterizingdiatonic and (hyper)pentatonicsets, respectively. From
theorems 2.2 and 3.6 we see that if Mc d is a diatonic set or if com-
pl(MCd) is a (hyper)pentatonic set, then Mcd E M(c), and com-
pl(MCd) E M(c). We now show that the connections among these
extremal, diatonic, and (hyper)pentatonic sets are still deeper.
Theorem 3.8. Let c - 8. Then the following are equivalent:
(1) Mcd E M(c), and compl(Mc, ) E M(c).
(2) M(c) is the set of diatonic sets in Uc.
(3) M(c) is the set of (hyper)pentatonic sets in Uc.
Proof. It is clear from theorems 2.1 and 3.5 that statement 1 holds
if c 0 (mod 4) and d = c/2 + 1, or possibly if c = 3 (mod 6) as the-
orem 2.1 does not include this case. First we will show that statement
1 does not hold if c = 3 (mod 6).
163
If c 3 (mod 6), then c is odd and, by theorem 3.6, M(c) = M
(c,(c - 1)/2). It follows that if compl(Mc,d) = Mc,c-d E M(c, (c - 1)/
2), then c - d = (c - 1)/2 and hence d = (c + 1)/2. But then
I spec(Mc,(c+1)/2) I = 2((c + 1)/2) - (c, (c + 1)/2) (lemma 2.14)
_ 2((c + 1)/2) - 1
- C.
165
"the multiplicity of clen j"):
#clen j = j + 1, for odd j < c/2
#clen j = c/2 - j + 1, for even j - c/2
#clen j = #clen c - j, for j > c/2
And for (hyper)pentatonic sets:
#clen j j - 1, for odd j < c/2
#clen j = c/2 - j - 1, for even j < c/2
#clen j = 0, for j = c/2
#clen j = #clen c - j, for j > c/2
The above may be verified by means of theorem 1.11. Now with
reference to fig. 11 (the diatonic case), one can see that after the lone
tritone (clen c/2) and 2 half-steps (clen l's) are in place, the filling in
of whole-steps (clen 2's) will yield-for odd clens-4 intervals of clen
3, 6 intervals of clen 5, etc., and-for even clens-3 intervals of clen
c/2 - 2, 5 intervals of clen c/2 - 4, etc. And similarlyfor the (hyper)-
pentatonic case. Gamer's (1967) work has helped us to understandthe
unique multiplicityproperty of deep scales (i.e., all dens of 1 through
c/2 appear in different numbers); from theorem 1.11 and the construc-
tion method discussed above we gain some insight about the patternof
those unique multiplicitiesfor one kind of deep scale-the diatonic-
and about the pattern of multiplicities (which are similarlyunique ex-
cept for nonzero dens) in its complement-the (hyper)pentatonic.
Interval circles
166
to the other intervals is immaterial; the circle is ME in any event.
However the circle of 2ds (7ths) is nontrivially even: it has two rare
intervals-half steps-and they are maximallyseparated. In summary,
all interval circles in M12,7 are maximally even; one of them and its
inversion are trivially so, the others nontriviallyso.
The following theorem shows that, quite generally, interval circles
in ME sets are themselves maximally even.
Theorem 3.11. Let c, d be fixed such that c - d > I - 1 and let
Mc.d = {Do, D,..., Dd 1j.Then, for allj, 0 - j d - 1, there exists
an Mcd = {Eo, E1 . ... } such that clen(Dj,,
),)
D(O+ = clen(Ej, Ei+ ), where the subscripts are reduced (mod d).
Proof. Since Mc d E M(c,d) there exists an m, 0 < m c - 1, such
that Mc.d = Jc, (theorem 1.5). Then
Jcd((j + 1) )- Jc(jI)
[cl(j +1) m]
d[ j d+ ]
m (mod c/).
Clearly clen(Djl D(+ )l) -clen(Ej, Ej+ ) (mod c), and clen(Dj,,
D(t+l)l) clen(Ej, Ej+ ). Hence
clen(Ej, Ej+ ) = nc + clen(Dj,, D(+I)/),
where n is a non-negative integer. Now we need only show clen(Ej,
- c - 1.
Ej+l)
By corollary 1.2, clen(Ej, Ej+l) = [], [1 + 1. Suppose d= c.
Then
167
(a) Circleof 3ds (b)M247
C - 1.
168
I dens of (Do, DI), (DI, D^2),(D21,D31).....(D121,Do)
1 l 2 2 2 2 2 l 2 2 2 2 22
2 3 4 4 3 4 4 3 4 4 3 4 4 4
3 5 6 5 6 5 6 5 6 5 6 5 6 6
4 7 7 8 7 7 8 7 8 7 7 7 8
5 9 9 9 9 10 9 9 9 10 9 9 9 10
6 11 11 1 111 1111 11 11 11 11
11 11 12
12 22 22 22 22 22 22 23 22 22 22 22 22 23
Second-order ME sets
The triads and seventh chords of tonal music are usually defined as
subsets of the diatonic set M12,7 (with allowances for inflection in the
case of the minor mode). If the usual diatonic set is itself conceived as
a subset of the 12-note chromatic universe, then triads and seventh
chords are subsets of subsets. A question naturally arises here: if, as
we have seen, the diatonic set is ME, are the familiar chords, its sub-
sets, also ME? But how shall we interpret this question? If we mean
"are the diatonic triads and seventh chords maximally even with re-
spect to the 12-note chromatic?" then the answer is clearly "no." Of
the three 3-note diatonic chords (major, minor, and diminished triads)
and four 4-note diatonic chords (the correspondences "three" with "3-
note," "four" with "4-note" are due to CV; see part 1 above), none
is ME in this sense. In fact no 3- or 4-note subset of M12,7 is ME in this
sense, for M12 3 = the augmented triad and M12,4 = the diminished-
seventh chord, neither of which is a subset of M12,7.
Now suppose we put the question differently: if the diatonic set is
maximally even with respect to the chromatic universe, are the tertian
chords maximally even with respect to the diatonic set? To answer this
question, think of the diatonic set as though it were chromatic. (M12,7
is, after all, ME, which is to say that, within the 12-note equal-
tempered template, it is the best approximation to 7-note equal tem-
perament.) From this perspective, triads and seventh-chords are
maximally even. At this level they are in fact instances of M7 3 and
M7 4, respectively-the best approximations to 3- and 4-note equal
temperament within the 7-note template.
169
This tri-level structureis illustratedin fig. 15, where the outer circle
represents the 7-note diatonic as small white circles within the 12-note
chromatic, and the inner circle represents the triad {0, 2, 4} and the
seventh-chord {0, 2, 4, 6} within the 7-note diatonic reconceived as
"chromatic." The interval spectrums of these chords with respect to
their seven-note context are given beneath the figure. Here the unit
dlen is the distance from one chord note to the next higher; unit clen
is the upwardstep in the seven-note set. Maximal evenness is evident
from the fact that each dlen is associated with two dens. We call such
sets second-orderME sets.
Thus the complete roster of first- and second-order ME sets con-
sists of the augmented triad, diminished-seventhchord, three diatonic
triads, and four diatonic 7th chords-in the view of many theorists,
precisely the set of chords capable of harmonic function in tonal
music. It is interestingto note in passing that the tertian 9th chords are
not second-order ME sets, but the usual pentatonic scale and its dia-
triad=20,2,4}
7t chord = 10,2,4,6}
a\ E
171
their implications for common tones among triads and 7th chords have
not been previously noticed, as far as we know.
The structure observed here is remarkable for what it shows re-
garding relationships between three nested levels of pc sets, which
might be labeled chromatic,diatonic, and harmonic. The celebrated
unique multiplicity property of the diatonic set with its consequences
for hierarchical structuring of transpositions of that set within the 12-
tone chromatic, as first observed, we believe, by Milton Babbitt
(1965), is present as well in triads and seventh chords, considered as
subsets within a seven-note world, where the consequences of that
property for hierarchical structuring of transpositions may reasonably
be regarded as an important feature of tonal harmony.
To generalize the above, chromatic universes which support dia-
tonic sets also support nested arrays of first-order, second-order, and
high-order ME sets that are deep scales with respect to their parent
collections. We suggest that the concept of diatonic system might well
include, as an essential feature, the hierarchical aspects discussed
here. Thus, it would encompass the diatonic system as defined by
Agmon (our diatonic set), plus one or more additional levels of struc-
ture including, in the usual case, the harmonic level.
Further investigations along these lines may be expected to yield
additional insights regarding tonal music as we know it, and to provide
a basis for invention of hypertonal musics in microtonal worlds.
172
Dembski, S. 1988. "Steps and skips from content and order: Aspects of a gener-
alized step-class system," Conference of the Society for Music Theory, Balti-
more, 1988.
Drabkin, W. 1980. The New Grove Dictionary of Music and Musicians, s.v. "dia-
tonic."
Forte, Allen. 1973. The Structure of Atonal Music. New Haven and London: Yale
University Press.
Gamer, C. 1967. "Some combinational resources of equal-tempered systems,"
Journal of Music Theory 11: 32-59.
Gauldin, R. 1983. "The cycle-7 complex: Relations of diatonic set theory to the
evolution of ancient tonal systems," Music Theory Spectrum 5: 39-55.
Morris, R. 1987. Composition with Pitch-Classes. New Haven and London: Yale
University Press.
Rahn, Jay. 1991. "Coordination of interval sizes in seven-tone collections," Journal
of Music Theory 35/1: 33-60.
Rahn, John, 1980. Basic atonal theory. New York: Longman.
Westergaard, P. 1975. An introduction to tonal theory. New York: W. W. Norton.
Yasser, J. 1932. A theory of evolving tonality, New York.
173