Volume 15, Number 5, October 2009
Copyright © 2009 Society for Music Theory
Joti Rockwell*Birdcage Flights: A Perspective on InterCardinality Voice Leading* 
REFERENCE: Richard Cohn, “A Tetrahedral Graph of Tetrachordal VoiceLeading Space” Music Theory Online 9.4
KEYWORDS: Voice leading, transformation, neoRiemannian theory, parsimonious graphs
ABSTRACT: This study explores connections between trichords and tetrachords from a voiceleading perspective. It begins with a definition of a voiceleading relation that draws from the work of Callender, Douthett, and Steinbach and uses a matrix to account for parsimonious voice leading. Given specific matrix values and chordtype constraints, networks arise that integrate both trichords and tetrachords, and both inter and intracardinality relations can be modeled by the resulting “birdcage” graphs.
Received March 2009
[1] After more than a decade of work in neoRiemannian and transformational theory, music theorists now have a strong handle on how to get from one consonant triad to another. Like a seasoned urbanite describing the quickest or most interesting route between two city locations, today’s trained music theorist can rattle off a bevy of moves one can make to get from one triad to the next.^{(1)} Recently, this knowledge of the harmonic terrain has been extended to include trichords other than major and minor triads, trichordal setclasses, seventh chords, tetrachordal setclasses, and chords and setclasses of other cardinalities.^{(2)} As theorists become increasingly skilled at mapping musical spaces, the pitch universe seems to be on the verge of being completely charted.
[2] With much of this theory resting on an epistemological foundation of voice leading, though, there are a number of challenges to the project of charting pitch space. Perhaps the most significant of these is the problem of characterizing voice leading between pitchclass sets of different cardinalities. In situations such as these, the question of accounting for an added pitch manifests itself as a mathematical uncertainty since a function, strictly speaking, cannot be used to model the motion both to and from each pcset. David Lewin, for example, calls attention to this issue in analyzing passages from Wagner containing both triads and seventh chords: “it would be impossible to express any of the voice leadings [of Lewin’s musical excerpts] as formal mathematical functions” (Lewin 1998, 67).
[3] One solution to the challenge of functionbased models of intercardinality voice leadings, proposed by Julian Hook as a “crosstype transformation,” involves considering chords as individual objects rather than collections of pitch classes.^{(3)} Functions can then be used to map a set of chords to and from another set of chords of different cardinality, provided that the respective chordsets are of equal size. This approach retains the algebraic clarity and coherence of intracardinality transformations by being a model of chord progression rather than voice leading.
[4] Incorporating intercardinality voiceleading relationships, however, can be a messy task. Joseph Straus has noted that with regard to a voiceleading graph, intercardinality moves among set classes can be “too numerous and intertwined” (Straus 2003, 63), and such moves have introduced a variety of complications into more recent spatial models of voice leading. In a later study by Straus, for example, the inclusion of sets of different cardinality results in setclass redundancies within his space, as well as rules prohibiting any type of move within the space that violates his definition of a shortest voiceleading path (Straus 2005, 57–59). For Tymoczko 2006, an intercardinality move manifests itself either as a shift from one multidimensional space to another space of different dimensions—a seemingly extreme form of spatial transportation—or as a move nearing a boundary of the higherdimensional space, in which case the intracardinality voiceleading properties of the lowercardinality chord are less revealingly organized.^{(4)} Perhaps this issue is, as Tymoczko 2007 argues, less a problem with the models than it is a reasonable interpretive and conceptual challenge to the theorist or analyst using them. Either way, the question remains as to whether musical spaces that incorporate both inter and intracardinality voice leading can match the cogency, suggestiveness, and analytic applicability of the successful singlecardinality models. The present study, which can be viewed as an exploration of subspaces of the more generalized studies of Straus 2005, Tymoczko 2006, and Callender, Tymoczko, and Quinn 2008, suggests that there may be some promise in this regard.
[5] This study examines connections between trichords and tetrachords from a voiceleading perspective. Drawing from the work of Callender 1998 and Douthett and Steinbach 1998, it begins with a definition of a voiceleading relation that uses a 2×2 matrix as a way of accounting for parsimonious voice leading for both inter and intracardinality moves. Given specific matrix values and chordtype constraints, networks arise that integrate both trichords and tetrachords. One such network proves to be a successful model for the “omnibus” progression, illustrating the pattern’s unique voiceleading properties. After formalizing these types of networks as “birdcage” graphs, this study finds all other such graphs for all types of parsimonious voice leading as given by the initial definition. Focusing on voice leadings as framed by particular chords, rather than chords as related by particular voice leadings, this study ultimately addresses the topic of intercardinality voice leading by answering the following questions: what parsimonious voice leadings allow for a trichordal set class and a tetrachordal set class to map both to themselves and to one another, and what are the set classes for which this situation can be met?
1. Modeling voice leading with the Pmatrix
[6] In a footnote to their investigation of parsimonious graphs, Jack Douthett and Peter Steinbach suggest a way in which their relation definition might be extended to accommodate pcsets of different sizes: motion from a larger pcset to a smaller one can be modeled as a “surjective” or “onto” map acting on each of the pitch classes in the larger set (Douthett and Steinbach 1998, 261). Example 1 illustrates the idea of such a model for two cases. In Example 1a, τ is the function moving G,B,D, and F to G,C,C, and E, respectively. This function contains no inverse: there is no singlevalued way of moving the C major chord to the G7 chord, since the pitch class C splits into two other pitch classes. In Example 1b, however, the function τ′ does have an inverse since the two pcsets have the same cardinality. The function is thus not just a surjection or “onto” mapping but a bijection, i.e., a mapping that is both “one to one” and “onto.”
Example 1. The τ function (click to enlarge) 
Example 2. Pmatrices (click to enlarge) 
[7] Given that composers move seamlessly within and among harmonies of various sizes, theories of voice leading do well to include both types of motion shown in Example 1. The present study thus models voice leading with a “parsimonious voiceleading matrix,” in which the pcsets under consideration need not be of the same cardinality. As Example 2 illustrates, this Pmatrix is of the form
u_{1}  u_{2}  
d_{1}  d_{2} 
[8] Pmatrices can be defined formally by invoking the surjective τ function from Example 1, thus allowing for “fusing” (and, indirectly, “splitting”) as introduced by Callender 1998. Definition 1 presents such a construction in which a chord X is greater than or equal to another chord Y in size, τ maps each pc in X to each pc in Y, and P indexes downward motion with the bottom row of the matrix and upward motion with the top row.^{(5)} Though this definition is limited to voice leading involving only one or two semitones, it can straightforwardly be extended to accommodate all sizes of voiceleading moves by expanding P into a 2 × z matrix, where the largest move is the size of z semitones. This generalized definition of voice leading is given as an Appendix and is not needed for the present purposes.
Definition 1: Given pcsets X and Y where the cardinality #X ≥ #Y, the two sets are related by a Pmatrix(written X
u_{1} u_{2} d_{1} d_{2} Y ) if there exists a surjection τ: X → Y and a set {x_{k}}
u_{1} u_{2} d_{1} d_{2} such that X\Y (the set of pcs in X that are not in Y) = {x_{k}}
#X1 k = 0 , and
d_{1} + d_{2} + u_{1} + u_{2}  1 k = 0
[9] The present voiceleading model, as illustrated by the first two examples and formalized in Definition 1, warrants two observations. First, Pmatrices are not unique to a given harmonic progression, since the voice leading can occur in a number of different ways. It is convenient to assume that the voice leading is as smooth as possible; that is, each voice in chord X moves to the nearest pc in chord Y.^{(6)} Given this constraint, P will still not necessarily be unique for any given progression (e.g., P for Example 2a could also be given as
1  1  
1  0 
[10] In addition to the flexibility of accounting for both inter and intracardinality moves, an advantage of characterizing voice leadings with P matrices is that while they can provide sums for total voice leading, they do not classify relationships by the sums alone. It seems overly general, for example, to consider the “hexatonic pole” transformation LPL to be the same type of voice leading as a T_{1} transformation on a triad.^{(7)} At the same time, Pmatrices are not so specific as to relate only to a single T_{n} or T_{n}I set class, as most of the common transformational labels do. An analytic benefit from this perspective is that musical works can be interpreted according to voiceleading equivalences—a type of motivic unity—when the harmonic palette is otherwise varied.^{(8)} For instance, Example 3 illustrates how
0  0  
2  0 
0  0  
2  0 
0  0  
2  0 
[11] As yet we have not accounted for motion from a smaller pcset to a larger one; since P involves the opposite scenario such motion is given by the “mirror” of P as given in Definition 2. A result of Def. 2 is that the relation given by P is noncommutative, since X P Y implies Y (MIR(P)) X. Furthermore, if a given nonzero P is the same as its mirror, some type of contrary motion is at work.^{(10)} Definition 2: For P = [12] With the voiceleading definitions now in place, various types of voiceleading parsimony can be explored, nine of which are proposed in Example 4. These types correspond to all Pmatrices in which the total voice leading is equivalent to three semitones or less. There is also one addition, type IX, which corresponds to the smoothest voice leading involved in a move between a tonic major triad and its dominantseventh chord (Examples 1a and 2a). Though this sort of motion is not especially parsimonious, strictly speaking, this voiceleading type appears in this study due to its frequency in the musical literature. The mirrors of each matrix in Example 4 are not included in the list, since they are grouped as part of the same voiceleading type. This consideration follows from the fact that common transformations correspond to both a Pmatrix and its mirror; e.g., the Ltransformation is represented by either

Example 3. Opening and Seconda parte, of Bartók’s Third String Quartet (click to enlarge) Example 4.
Nine types of parsimonious voice leading (click to enlarge) 
2. “Birdcage” graphs
[13] As an initial case, consider the second type of voice leading from Example 4, in which two voices move in contrary motion, each by a semitone. This type stands out from the others in that it is the only one with mirror symmetry. Allow also the constraint in which the two chords involved are either minor triads or dominant seventh chords. We thus have P =
[14] If we represent this scenario as a “parsimonious graph,” in which vertices represent pcsets and edges represent voiceleading relations, we obtain the figure given in Example 6, which this study names a “birdcage graph.” Some familiar structures from transformational theory are immediately apparent in this graph: each of the four outer triangles connecting the minor triads represents a subgroup of the neoRiemannian LP group (a cyclic group of order 3 which can be represented by its generators as 〈LP〉 or 〈T_{4}〉), while the inner squares connecting the seventh chords can also be considered as a subgroup of the group of transformations on dominant seventh chords (a cyclic group of order 4 given by 〈T_{3}〉 or, with the transformations of Douthett and Steinbach, 〈L_{2}^{*}P_{2}^{*}〉 (Douthett and Steinbach 1998, 249–252)). The union of the pc’s of any triangle connecting minor triads is a hexatonic set, while the union of the pc’s of any square is an octatonic set.^{(11)} [15] What is perhaps most interesting about this graph, though, is that the only case in which the Pmatrix

Example 5. Possibilities for P = Example 6. Graph for X,Y ∈ {T_{n}{0,4,7,10},
Example 7. Threecolor subgraph for X,Y ∈
(click to enlarge) 
[16] The birdcage graph of Example 6 gains particular relevance when we use it to model the “omnibus” progression of the type explored by the theorist Georg Vogler around the turn of the nineteenth century.^{(14)} There are a number of ways of discerning a group structure from the graph, most of which involve combining the cardinalitychanging transformation with some of the known groups that act on triads and seventh chords.^{(15)} For the present purposes, though, we will restrict ourselves to the cardinalitychanging Pmatrix transformation discussed above, labeled here as α, along with transposition. As Example 8 illustrates, beginning with any of the minor triads and repeatedly applying an <α, T_{9}, α> sequence of transformations induces one of three possible “omnibus subgraphs,” around which the omnibus progression follows a clockwise path.
Example 8. Omnibus subgraph 
Example 9. Brahms, third symphony, Poco Allegretto, measures 150–154 
[17] Though many omnibus patterns can be heard in commonpractice music, composers can navigate the space of
Example 6 without recourse to a continuously descending chromatic line. Brahms, for example,
can be heard to utliize the symmetry of this space toward the end of the third movement of his third symphony. As
Example 9 illustrates, E7 and
[18] The combination of inter and intracardinality relationships in Example 9 addresses an issue arising in the multiset voiceleading approach presented in Tymoczko 2006 and Callender, Quinn, and Tymoczko 2008. Their work renders collections of pcsets such as those of
Example 6 as one of their “quotient spaces,” in which some combination of five types of equivalence is applied to an ndimensional space with real coordinates (^{n}), with the coordinates designating ordered
pitch sets of cardinality n. Chord spaces form as a result of octave and permutation equivalence (e.g., “gluing together”
[19] The issue is n. In the case of intracardinality moves, chords relate to one another in the same orbifold, and the geometry of the orbifold can lead to observations about both the chords and the voice leadings involved. One can define transformations as paths in the space, thus combining geometrical and grouptheoretic approaches (indeed, group actions have already been used to create the orbifold); one can use the geometry of the space to measure voiceleading distance; and one can, as the present study does, render points and paths in the space as a mathematical graph, thus focusing solely on chords and voiceleading relationships. In the case of intercardinality moves, though, the two types of chords are in two different orbifolds, each of which has a different number of dimensions.
[20] The authors present three ways of handling this issue. First, glue together the various spaces by means of “cardinality equivalence” and use the resulting “C space” to model intercardinality progressions. Such an operation distorts the geometry of the orbifolds, though, such that the resulting space is impractical for examining voice leading (Callender, Quinn, and Tymoczko 2008, supporting materials, 14). Second, treat the smaller chord as a multiset in the higher dimensional orbifold. In many cases, this is a productive approach. The omnibus progression, for example, achieves a particularly revealing interpretation this way: the chords of
Example 8 appear in the 4dimensional orbifold as the sequence
[21] With the structure of Examples 6–9 in mind, Definition 3 presents a formal conception of a birdcage graph. Essentially, the definition states that for a graph to be a birdcage graph, the voiceleading relation given by P must allow both cardinalitychanging and cardinalitypreserving moves. It also rules out the voiceexchange relation in which a pcset is mapped to itself, such as {0,1,4,7}→{1,0,4,7} for P =
1  0  
1  0 
Definition 3: Given a Pmatrix and a graph with vertex set V = {v: v ∈ X ∪ Y} where X and Y are T_{n}I set classes of cardinality 4 and 3, respectively, and edge set E = {e: e ∈ (v_{a}, v_{b})} such that v_{a} P v_{b}, the graph is a “birdcage” graph if there exist edges e_{X}, e_{Y}, and e_{X,Y} such that:e_{X} = (x_{1}, x_{2}) where x_{1} ≠ x_{2} and x_{1}, x_{2} ∈ X
e_{Y} = (y_{1}, y_{2}) where y_{1} ≠ y_{2} and y_{1}, y_{2} ∈ Y
e_{X,Y} = (x,y) where x ∈ X and y ∈ Y
[22] Example 10 visualizes the definition with a birdcage “setclass pseudograph,” which contains two vertices connected by an edge and a selfloop (hence the prefix “pseudo”) at each vertex. The pseudograph of any combination of X, Y, and P that satisfies Def. 3 will have the form of Example 10, since both set classes have the ability to selfmap and map to one another via P. By including T_{n}I set classes, Def. 3 leaves open the possibility for birdcage graphs to have multiple components that partition the tetrachordal and trichordal families, provided that each component connects at least two chords of each cardinality. With the precise definition in place, two questions arise: first, do birdcage graphs exist that include chords other than minor triads and dominant sevenths if P =
[23] Since birdcage graphs involve T_{n}I set classes, the most logical initial move in searching for graphs in addition to that of Example 6 is to invert each of the chords in the example. This transformation produces an identical structure that relates major triads and halfdiminished seventh chords. Example 11 illustrates such a graph along with a corresponding progression labeled here as a “Ziehn Inverted Omnibus” after Bernhard Ziehn, who included this progression in his Canonical Studies.^{(18)} Since Example 6 and Example 11 contain all the chords of each transposition class for X = [0258] and Y = [037], they combine to form the complete birdcage graph for the two set classes. [24] The harmonic progression of Example 11 is not employed nearly as frequently in the musical literature as the uninverted form, most likely because it is considerably more difficult to integrate within a tonal context. Telesco (2001, 134) suspects that “at least some instances exist,” but even Wagner, whose interest in both parsimonious voice leading and the dramatic potential of the half diminished seventh chord would in principle lead him to such a progression, seems not to use it in any extended capacity. It is absent, for example, in Gauldin’s 2001 and 2004 studies of doublesemitonal voice leading and wedge progressions in Wagner’s music. Nonetheless, Hunt 2007 has traced a number of motivic correspondences in the Ring cycle that utilize the

Example 10. Birdcage setclass pseudograph
Example 11. a) Graph for X,Y ∈ {T_{n}{0,3,6,10}, T_{n}{0,4,7}}
Example 12. Nearbirdcage for X,Y ∈ {T_{n}{0,3,7,10}, T_{n}{0,4,8}} 
[25] Another possibility for a birdcage graph would invoke the minor seventh chord’s abilities to move parsimoniously among its members. The resulting graph connects these seventh chords with augmented triads, though it is technically not a birdcage graph since there are no two augmented triads that can be related by P. Example 12 illustrates this graph, which has fewer edges and vertices in part because of the symmetry of its pcsets: augmented chords have only four unique transpositions, while the inversionally invariant minor sevenths (unlike dominant and halfdiminished 7ths) are not related to fully diminished seventh chords by a semitone, resulting in fewer P connections to other members of their own set class.
[26] Computing all voiceleading possibilities for P =
1  0  
1  0 
[27] Example 13a presents a birdcage graph that includes minor triads and {0,1,5,7} transpositions, which lead to a rather different structure than that of Examples 6, 11, and 12. Unlike the previous examples, this graph is planar and consists of two components (four for the complete birdcage graph when all [037] and [0157] members are included), and it is possible to walk through each of the vertices of each component graph without crossing the same vertex twice. As the musical realization of
Example 13b suggests, though, such twelvechord “Hamiltonian” paths lack the voiceleading consistency of the omnibus progression, despite the fact that they can be built on a descending chromatic bassline.^{(21)} The subgraph realization of
Example 13c is more like the omnibus progression since it utilizes intercardinality involutions, but the lack of the chromatic voice exchange (e.g., the
Example 13. a) Graph for X,Y ∈ {T_{n}{0,1,5,7}, T_{n}{0,3,7}} (click to enlarge) 
Example 14. a) Opening to Berg’s Piano Sonata, Op. 1 (click to enlarge) 
[28] The desultory upper voice leading of Example 13b and the static nature of Example 13c would appear to make such patterns less desirable to composers than would the omnibus progression, and indeed they are not extensively used in the literature. Nonetheless, the graph of Example 13a proves useful in considering the music of Alban Berg, whose use of both parallel and contrary (“wedge”) semitonal voice leading is welldocumented (Headlam 1996, DeVoto 1991, and Schmalfeldt 1991). In Berg’s case, steadily progressing chromatic lines do not tend to maintain setclass consistency as they do for the omnibus progression, such that voice leading is often more of a unifying factor than harmonic type. Notwithstanding local harmonic variety, set classes [037] and [0157] seem to play a consistently important role in his works, most likely due to their usefulness in contexts that combine tonality and atonality. Example 14 gives two important instances in which Berg utilizes the parsimonious relations between these set classes.^{(22)} In the opening of his Piano Sonata, Op. 1, numerous instances of descending doublesemitone motion give way to the mirror voice leading for a V→i cadence in B minor. The opening tetrachord {11,1,6,7}, which is a particularly significant sonority in this work and others (Schmalfeldt 1991, 92–94), connects to B minor, the key of the sonata, via contrary double semitonal motion. As Example 14b shows, this same motion along with an [0157] selfmapping generates a motive that appears frequently in the Altenberg Lieder.^{(23)} This selfmapping corresponds to Lewin’s RICH transformation, which maps an ordered pcset to its retrograde inverted form that begins with the initial set’s final two pcs (1987, 180–188).
[29] We have now exhausted all the birdcage possibilities for the typeII matrix (P =
[30] Since Pmatrices other than
[31] Example 16a gives a section of the birdcage graph for [0124] and [013] using the Pmatrix
[32] The graph of Example 16 recalls Bartók’s dense motivic chromaticism as analyzed in Example 3, which illustrated how the Third String Quartet begins with an [013] selfmapping via
[33] Example 18a gives the first of the two birdcage graphs corresponding to P =

Example 15. All birdcage possibilities for all 9 voiceleading types
(click to enlarge) Example 16. Slice of graph for X∈[0124], Y∈[013], P = (click to enlarge) Example 17. Opening of Bartók’s Sixth String Quartet (click to enlarge) Example 18. a) Graph for X∈[0247], Y∈[037], P = 
[34] Example 19a presents the latter of the two birdcage graphs that incorporate consonant triads and majorminor/halfdiminished seventh chords. The graph partitions the four transposition classes of the 48 chords into wholetonerelated sets, due in part to the wholetone voice leading that generates it. While the components of this graph could be represented as circles rather than gearlike structures, the current rendering makes it easier to visualize how “major” and “minor” chords are each connected: major triads and their tetrachordal supersets occupy the outer rings of the graph, while minor triads and their tetrachordal supersets occupy the inner rings.
Example 19. a) Graph for X∈[0258], Y∈[037], P =

Example 20. a) opening and
(click to enlarge) 
[35] Chained 5vertex subgraphs of Example 19a give a useful family of jazz chords over a common bass note. As
Example 19b illustrates, the minor triads at the edges of these subgraphs form incomplete “add6” and “9” chords over the root of the two “major” chords in the segment.^{(25)} The five vertices thus comprise a set of possible chords that a pianist can play with her right hand over a lefthand foundation corresponding to the major triad in the set. The union of these five chords forms a 13 chord in which the 9th is included. As Example 20 illustrates, Miles Davis’s “Freddie Freeloader” (1959) begins by moving between the outer chords of this collection, and the beginning of his trumpet solo suggests a more complete path. The great staff below the trumpet line in
Example 20b charts chords that Miles Davis’s melodic tones initiate, and the end of the phrase skips through
[36] The results of this study can be summarized as follows:
 The original birdcage graph that models the omnibus is the only one involving splitting and fusing in the strict sense. In other words, it is the only graph in which the changing trichordal pc splits into two different pcs (or, conversely, the only one in which two voices in the tetrachord move to create the fused pc in the trichord). Thus, the omnibus progression and its inverse are unique from a voiceleading perspective.
 None of the set classes in any birdcage graph has inversional symmetry, which means that each complete graph has 48 vertices representing the four transposition classes of the trichords and tetrachords.
 Despite the absence of these set classes, all pcsets that can form birdcage graphs are minimal perturbations of inversionally invariant sets. In other words, each of the sets in a birdcage graph can be connected to an inversionally invariant set by a single semitonal move. For example,
{0,2,5,8} {0,3,7}
{0,2,6,8};1 0 0 0
{11,3,7}; and {0,1,6}0 0 1 0
{11,1,6}.0 0 1 0  The high degree of interconnectedness for
voice leadings (birdcages 3–10) results in graphs that have only one component, while the limited voiceleading possibilities of the other two voiceleading types result in various partitions of the 48 chords.2 0 0 0  In providing the conditions under which birdcage graphs exist, this study demonstrates how pcsets can map to members of their own set class, a phenomenon that has been privileged in the neoRiemannian literature.^{(26)} These results can help in judging the extent to which diatonic collections are indeed “overdetermined.” The consonant triad, for
example, has one degree of “hexpole”type connectedness (e.g.
{0,3,7} , while [026] has two (e.g. {0,2,6}
{11,4,8})2 0 1 0
{11,3,5}, and {0,2,6}1 0 2 0
{11,1,7}).1 0 2 0  Birdcage graphs can be used in the analysis of both tonal and atonal literature. Still, as with the other spatial models that relate to this study, the full extent to which birdcage graphs are musically practicable will be borne out through further analytic study.
 The mathematical concepts employed here, namely surjective mappings and graphs, intersect but are not coextensive with grouptheoretic and geometrical approaches to music theory. Again, analysis, perhaps along with composition and studies in the psychology of music, will be critical in determining the degree to which the various mathematical metaphors hold.
Appendix: Generalized VoiceLeading Definition
Definition 1a: Given pcsets X and Y where the cardinality #X ≥ #Y, the two sets are related by a Pmatrix
if there exists a surjection τ: X → Y and a set {x_{k}}
u_{1} u_{2} ... u_{z} d_{1} d_{2} ... d_{z} such that X\Y = {x_{k}}
#X1 k = 0 , p_{sum} =
p_{sum1} k = 0 (d_{n} + u_{n}), and
z ∑ n = 1
Joti Rockwell
Pomona College
Department of Music
340 N College Ave, Claremont, CA 91711
Works Cited
———. 1998. “Square Dances with Cubes.” Journal of Music Theory 42, no. 2: 283–296.
———. 2003. “A Tetrahedral Graph of Tetrachordal VoiceLeading Space.” Music Theory Online 9, no. 4.
Headlam, Dave. 1996. The Music of Alban Berg. New Haven: Yale University Press.
Hook, Julian L. 2002. “Uniform Triadic Transformations.” Ph.D. dissertation, Indiana University.
———. 2002. “Uniform Triadic Transformations.” Journal of Music Theory 46, no. 1 and 2: 57–126.
———. 1998. “Some Ideas About VoiceLeading between Pcsets.” Journal of Music Theory 42, no.1: 15–72.
———. 2005. “Voice Leading in SetClass Space.” Journal of Music Theory 49, no. 1: 45–108.
Tymoczko, Dmitri. 2006. “The Geometry of Musical Chords.” Science 313, no. 5783 (7 July): 72–74.
Yellin, Victor Fell. 1998. The Omnibus Idea. Warren, MI: Harmonie Park Press.
Footnotes
* I would like to thank Richard Cohn, Joseph Straus, Dmitri Tymoczko, and the anonymous reviewers at
MTO for helpful feedback on this research. I would also like to thank Jack Douthett for providing detailed comments on an earlier draft.
Return to text
1. Of these possible triadic moves—usually referred to as “transformations,” “operations,” or “voice leadings”—the most commonly cited are the canonical twelvetone operators of transposition and inversion, along with the more recently formalized neoRiemannian operations of L (Leittonwechsel), P (Parallel), and R (Relative). These five moves hardly exhaust the possibilities, though, since there are 620 × 10^{21} ways (i.e., 24 factorial) of mapping the set of 24 consonant triads onto itself. This unwieldy group of possibilities is tamed by Hook 2002,
which integrates a variety of commonly referenced triadic transformations into his “UTT” formalism.
Return to text
2. See Callender, Tymoczko, and Quinn 2008, Tymoczko 2006, Straus 2005, Cohn 2003, Gollin 1998, and Childs 1998.
Return to text
3. Hook 2002b, further expanded in Hook 2007.
Return to text
4. Headlam and Brown (2007) have raised a similar issue regarding the geometric position of relatively smallcardinality pcsets in higher dimensional spaces. A later section of the present
study addresses the topic of Callender, Quinn, and Tymoczko’s multidimensional spaces in more detail.
Return to text
5. Readers may wish to skip the latter portion of Definition 1, since an understanding of the material to follow is not contingent upon knowing the specifics of how τ is defined.
Return to text
6. Straus 2003 calls this the “smoothest” or “most efficient” voice leading. Cohn 1998 considers voiceleading efficiency as a function VLE which sums all the minimal voiceleading distances in a given progression. In terms of Pmatrices, the sum of all the matrix values for X→Y
Return to text
7. Examples of these neoRiemannian and twelvetone transformations are
Return to text
8. Andrew Pau (2007) has shown evidence of this type of voiceleading unity in a number of works by Schoenberg and Berg.
Return to text
9. As this second phrase is more complex than the first, other hearings are eminently plausible, including
1  0  
2  0 
2  0  
1  0 
Return to text
10. Admittedly, the notion of a “mirror” is a way of working around the fact that the surjective nature of τ precludes an inverse function for intercardinality mappings. A common solution to this issue is to work with multisets, in which case voice leadings are always bijective. Analytically, however, multisets pose the problem of “phantom doublings” in which the size of a chord cannot be ascertained from the number of pcs it possesses. Splitting and fusing are viable musical concepts that avoid such a problem. This study thus
explores the surjective model, notwithstanding the fact that in the case of splitting, the frequently used Lewinian metaphor between musical motion and mathematical functions breaks down.
Return to text
11. I would like to thank Jack Douthett for this observation about pc unions.
Return to text
12. In graphtheoretic terms, this is not a proper coloring because two or more edges of the same color are adjacent to one another for the intracardinality cases. The familiar problem in graph theory of finding minimal edge colorings is thus similar but not identical to the question of finding unique voiceleading types within a voiceleading graph.
Return to text
13. This function is the “S_{2}” split transformation of Shimbo 2001 (described in Hook 2002a, 117). Hunt 2007 refers to it as a “Leittonwechsel split” (“S^{*}_{L}”), since a neoRiemannian L transformation is contained within it. Hunt’s conception is thus similar to that of Hook (2002b, 117–118 and 2007, 2–3), who connects the same types of chords (e.g., c→A7) through the combination of L and a crosstype “inclusion transformation” that maps [037] triads to their unique [0258] supersets. Unlike the other authors, Hook avoids defining his crosstype transformation as an involution, since having a transformation acting on a set of 48 triads and seventh chords leads to algebraic complications (2007, 5, footnote 8).
Return to text
14. Hence the motivation for the term “birdcage graph.” For an examination of Vogler’s relationship to this progression, as well as numerous musical examples invoking omnibus progressions, see Telesco 1998 and Yellin 1998.
Return to text
15. See Hook 2002b, 116–118 and Hook 2007, 1–3, both studies of which argue that the examination of the group actions he defines quickly leads to realizations about the omnibus progression’s voiceleading efficiency. The present study has made the opposite point: exploring parsimonious voice leading between triads and seventh chords leads in this case to an algebra of harmonic transformation for which Hook’s crosstype and GIS homomorphism theory is the foundation.
Return to text
16. Callender, Quinn, and Tymoczko (2008) demonstrate how the omnibus progression forms a line in their 4 dimensional chord space, while also relating the progression to the geometry of Knets (supp. mat., 16–19 and 39–44). The 2dimensional graph in Tymoczko’s Fig. 2 (2006, 73), though, suffices to illustrate the phenomenon of reflection via contrary semitonal motion.
Return to text
17. In this space, {0,3,3,7} and {11,3,3,8} are the triadic multisets that preserve the contrary semitonal motion between a and c. Not all examples afford such a possibility, since if one is interested in doublesemitonal motion for dyads, the 2dimensional space needs to be used; e.g.,
Return to text
18. Ziehn 1912, as described in Yellin 1998, 9–13.
Return to text
19. The voiceleading calculations were done with a computer program written in Python.
Return to text
20. Voiceleading efficiency for [0157] selfmappings is in keeping with longheld observations in neoRiemannian theory that link a chord’s ability to parsimoniously selfmap to its closeness to inversional symmetry, transpositional symmetry, and/or equal division of the octave. See Cohn 1997, 1–6; Callender 1998, 221; Tymoczko 2006; and Callender, Quinn, and Tymoczko 2008, supp. mat.,10–13.
Return to text
21. The descending chromatic line of Example 11b can be completed, thus rendering the overall progression as a cycle, by moving from d to
Return to text
22. The one type of relation not given in Example 14, that of parsimonious voice leading between consonant triads, belongs more to the music of Berg’s predecessors than it does to his compositions.
Return to text
23. DeVoto 1966 identifies this motive, labeling it α, in three of the five songs. Schmalfeldt (1991, 93–94, footnote 26) notes that it contains “interlocking presentations” of [0157].
Return to text
24. Alternatively, following Straus 2008 and Lewin 1987, one can draw serial mappings between melodic fragments of equal cardinality. Excluding the first two notes, the passage would then constitute an chain of four overlapping trichords related by inversion, retrograde inversion, and retrograde inversion: (8,9,11) →(11,10,8)→(8,6,5)→(5,4,2).
Return to text
25. The add6 chord lacks a chordal fifth, while the add9 chord lacks a chordal third. Jazz players routinely omit chord tones according to taste, although leaving out the chordal fifth is far more common than omitting the chordal third, which plays a more important role in defining the sonority. For studies that explore parsimonious voice leading in jazz, rock, and pop music, see Strunk 2003, Santa 2003, and Capuzzo 2004.
Return to text
26. See Cohn (1997), who attributes the “overdetermined” quality of consonant triads to their ability to selfmap via parsimonious voice leading.
Return to text
Copyright Statement
Copyright © 2009 by the Society for Music Theory. All rights reserved.
[1] Copyrights for individual items published in Music Theory Online (MTO) are held by their authors. Items appearing in MTO may be saved and stored in electronic or paper form, and may be shared among individuals for purposes of scholarly research or discussion, but may not be republished in any form, electronic or print, without prior, written permission from the author(s), and advance notification of the editors of MTO.
[2] Any redistributed form of items published in MTO must include the following information in a form appropriate to the medium in which the items are to appear:
This item appeared in Music Theory Online in [VOLUME #, ISSUE #] on [DAY/MONTH/YEAR]. It was authored by [FULL NAME, EMAIL ADDRESS], with whose written permission it is reprinted here.
[3] Libraries may archive issues of MTO in electronic or paper form for public access so long as each issue is stored in its entirety, and no access fee is charged. Exceptions to these requirements must be approved in writing by the editors of MTO, who will act in accordance with the decisions of the Society for Music Theory.
This document and all portions thereof are protected by U.S. and international copyright laws. Material contained herein may be copied and/or distributed for research purposes only.
Prepared by William Guerin and Mitch Ohriner, Editorial Assistants
Updated
12 November 2009
Number of visits: