peter.beimgraben@b-tu.de
Escola Superior de Música de Catalunya
Departament de Teoria, Composició i Direcció, Spain
thomas.mamuth@gmail.com
Quantum Tonality: A Mathemusical Playground
Abstract
We adopt some basic ideas on quantum-theoretical modeling of tonal attraction and develop them further in an alternative direction. Fitting Gaussian Mixture Models (GMM) to the Krumhansl-Kessler (KK) probe tone profiles for static attraction opens the possibility to investigate the underlying wave function as the stationary ground state of an anharmonic quantum oscillator with a schematic Hamiltonian involving a perturbation potential. We numerically verify the fulfilment of the associated stationary Schrödinger equation and also inspect its excited states as a solution basis for the corresponding time-dependent Schrödinger equation. With their help we calculate the temporal evolution of any initial state. As an example, we study the dynamics of transpositions of the stationary KK wave function across Regener’s line of fifths. This offers potential models for dynamic tonal attraction and also for the behavior of deflected key profiles within the Hamiltonian dynamics of a given tonality.
Keywords:
Quantum Theory Music Theory Pitch Class Profiles Krumhansl-Kessler Probetone Profiles Gaussian Mixture Models Quantum Harmonic Oscillator Schematic HamiltonianTo appear in the Proceedings of the MCM Conference 2024, Springer Verlag, Cham.
1 Introduction
In the present paper we experiment with a specific concept of tonality that enters music theory from studies in cognition on the one hand and computer aided analysis on the other. The psychological Krumhansl-Kessler (KK) probe tone profiles [10] for the major and minor keys describe static tonal attraction of the various pitch classes in a tonal context and thereby reflect traces of enculturation within the Western musical mentality [9], [18]. Related profiles also result from statistical analyses of pitch class distributions in musical corpora and several authors have discussed connections between both kinds of data, e.g. [2], [9]. In a more refined way, pitch class profiles are also used to represent the (fuzzy) pitch class content of local score segments and the investigation of their Discrete Fourier Transforms (DFT) provides valuable analytical insights through the tonal interpretation of the phases of prominent Fourier coefficients (see [1], [7], [20]). The KK profiles have also been interpreted as manifestations of pitch class hierarchies. For example, in his book Tonal Pitch Space, Fred Lerdahl [11] adopted the idea of embedded pitch hierarchies from Diana Deutsch und John Feroe [4], and analyzed tonal chord progressions in terms of such hierarchies. Moreover, melodic probe tone profiles and pitch class transition probabilities play an important role for dynamic tonal attraction [9], [19]. These remarks indicate that pitch class profiles enter music theory in connection with various approaches to the investigation of tonality. Upon this wider background it is desirable to understand the implications of a more recent approach to the mathematical interpretation of the KK profiles, proposed by Reinhard Blutner and Peter beim Graben in [5], [6] using ideas from quantum theory.
Starting point of the present study is the observation in [14, Fig. 1] that the deformed quantum cosine model over the continuous circle of fifths [6] can be equivalently described by a Gaussian wave function over Regener’s line of fifths as configuration space [15], retaining the probe tone positions along the circle of fifths up to enharmonic equivalence, such that the KK profiles emerge as mixed quantum states, i.e. as weighted sums
(1) |
of three transposed instances of a Gaussian probability density function , where refer to the positions of the three components of the tonic triad. This finding leads to two important consequences. (i) it links the quantum music approach to statistical machine learning and data science [16], and (ii) it allows the application of the quantum theory of harmonic and anharmonic oscillators [17],[12].
2 Quantum Wave Functions for Gaussian Mixture Models
This section summarizes the results from a longer companion article by the authors of the present submission [8]. First we recall that a Gaussian wave packet
(2) |
with energy
(3) |
and variance solves the stationary Schrödinger equation of the harmonic oscillator
(4) |
for the ground state energy [17].
Next, we consider a generalized Gaussian mixture model (GMM) [Eq. (1)] of the form
(5) |
where
(6) |
is a normal distribution density with variance , whose mean is shifted to a tone position in each corresponding summand for the given tonality, altogether constituting a tonal context [6]. The coefficients are the respective weights in the convex linear combination (5) with . Such GMM play an important role in the domain of statistical machine learning [16]. Here, they serve as a mathematical model for the schematic traces of musical enculturation [9], [18].
In the companion article [8], we show that there exists a perturbation of the harmonic oscillator potential such that the ground state solution (for the same energy eigenvalue ) of the anharmonic stationary Schrödinger equation [12]
(7) |
has the probability density function of the GMM (5). Without proof we mention here, that the perturbation potential is given by the formula
(8) |
where and the parameters and are the defining ingredients of the GMM (c.f. equations (3, 5) and (6)). The differential operator
(9) |
is referred to as the schematic Hamiltonian associated with the GMM (5) reflecting a statistically acquired tonal schema [9], [18].
3 Fitting Gaussian Mixture Models for Static Tonal Attraction
In the present paper we explore these results for the case of two concrete examples, namely a GMM (1) for the KK probe tone profiles in major and minor [10]. Table 1 lists both data sets (major and minor) in a normalized form:
The note names in the line-of-fifth order (with added to the left side with the same values as for ) occupy the equidistant positions , centered around the note at the origin from the original quantum model over the circle of fifths. We write for the position of the note at the real line of fifths [15]. The original values are scaled such that the norms of the linear interpolations yield unity. We fit several GMM onto the KK data in analogy with the approach in [6], i.e. we choose GMM of the form (1) with different constraints upon the parameters.
The choice of the circle of fifths interval motivates the inheritance of the ground state energy
(10) |
from the Lerdahl interpolation in [3].111 Note that we are able to confirm Lerdahl’s speculation about a Hamiltonian principle of musical least effort [11] in the present framework, as the ground state energy can be obtained from a corresponding variational principle. In order to fit the most general GMM to the KK data we have five parameters: and ( results from the convexity constraint ). The locations of the three mean values are musically constrained through the assumption of the formative role of the tonic triads, i.e. for the major context and for the minor context. Beyond this it is also interesting to compare these choices with “musically ignorant” fits to see how far they deviate from the musically motivated ones.222 The authors are acutely aware of the danger of overfitting, in particular as the shape of evokes a connection between the elephant of Antoine de Saint-Exupéry’s The Little Prince and John von Neumann’s elephant: “With four parameters I can fit an elephant, and with five I can make him wiggle his trunk.” [13]
The results, displayed in Fig. 1, indicate that in the case of the major key profile there is no significant difference between the two fits. In the case of the minor key profile, however, there are noteworthy differences in connection with the raised scale degrees and which deserve a separate investigation. For the scope of the present paper we focus on the case of the major key.333 Fitting with a higher energy parameter one can even achieve a correlation of , unfortunately, the match for the raised leading tone is not yet convincing.
4 Wave Functions of Static Tonal Attraction as Schematic Stationary Ground States
We proceed with the major GMM (1) for tone positions from the top left cell of Fig. 1, with , and defining our candidate for the associated schematic ground state wave function as .
Our next goal is to verify that is indeed the stationary ground state of the schematic Hamiltonian (9) by solving the eigenvalue problem of the operator . For this purpose, it is useful to take advantage of the Hermite basis of the harmonic oscillator, described by (11).
(11) |
where denotes the Hermite polynomial of order . These Hermite states are solutions of the unperturbed Schrödinger equation (4) for the eigenvalues , i.e.
(12) |
When we apply the schematic Hamilton operator (9) to a superposition state
(13) |
with coefficients , we obtain
The coefficients for eigenstates of the schematic Hamilton operator form eigenvectors of the infinite matrix where
(14) |
(for details see the corresponding companion paper [8]). The potentials for the harmonic oscillator and its perturbation in our concrete example (see (8)) are shown in Fig. 2:
For a numeric validation of the schematic Schrödinger equation, we calculate eigenvectors and eigenvalues of in growing (but finite) number of dimensions by truncating the infinite matrix after rows and columns: .
The upper segment of Fig. 3 shows list plots of the first (top left) and the first (top right) coordinates of with respect to the Hermite basis (12) and compares them with the coordinates of the “smallest” eigenvectors of and , respectively (i.e. the eigenvectors with the corresponding smallest eigenvalues). These eigenvalues ( and ) are close to the Lerdahl ground state energy (10), while one can still observe a clear deviation between the coefficients in the case of 11 dimensions (top left side of Fig. 3) the deviation already gets smaller in the case of 16 dimensions (top right side of Fig. 3). The plot in the lower segment of Fig. 3 illustrates the fulfillment of the Schrödinger equation. The dashed curve shows the wave function , i.e. our predicted ground state stretched by the predicted energy eigenvalue , while the continuous curve shows the image of the smallest eigenstate under the action of , which is the approximation of the schematic Hamilton operator in terms of the truncated matrix . For growing the little oscillations of the Hermite components become smaller and smaller and their superposition approaches the predicted wave function.
5 Excited States of the Schematic Hamiltonian
In this section we explore an immediate consequence of the fact that we are in the possession of a Hamilton operator. Apart from the ground state we have an infinite sequence of higher energy eigenvalues and an infinite basis of corresponding excited stationary states, which form an orthonormal basis for the space of square-integrable complex functions over the line of fifths . In order to explore them, we may inspect their approximations from the sequence of eigenvectors of the truncated matrices . Figure 4 suggests that — for growing — the eigenvalues of converge in quadratic mean the eigenvalues of from above, too.
Mathematically, we will now use the excited states in order to calculate the time-development of given initial states under the schematic Hamiltonian according to the time-dependent Schrödinger equation
(15) |
6 Time Development of Deflected Key Profiles
The excited states of the schematic Hamiltonian , denoted , form an orthonormal basis of the space of complex square-integrable functions on . The stationary ground state is the KK profile of static tonal attraction.
Suppose we have a representation of a wave function
(16) |
whose coefficients are obtained by the orthogonal projections
(17) |
Then, the solution of the time-dependent Schrödinger equation (15) is universally given by
(18) |
with the eigen energies of the schematic Hamiltonian (9).
Assuming that represents the ground state of a given tonality (here major), it seems reasonable to associate other (major) tonalities with transpositions of . However, regarding the underlying dynamics of the major tonality such transpositions are not autonomous representations of other tonalities at all. While is stationary, for is not. The coherent states of the (unperturbed) quantum harmonic oscillator may serve as prime examples for our situation. They are the quantum-theoretical analogues for a deflected pendulum. We will therefore baptise the wave functions (in the role as initial states) as deflected key profiles. In order to explore their time developments we need to calculate the orthogonal projections
(19) |
Figure 6 gives a simultaneous flipbook-like trace (of the absolute values) of two time developments, namely of the fifth deflected wave functions in sharp-ward and in flat-ward direction. Both wave functions show the tendency to eventually arrive at the location of the stationary ground state. But in the details they behave quite differently and they never seem to match the precise shape again. From the numeric computation in 30 dimensions alone we cannot tell whether the deviations arise completely from the finite-dimensional approximation. A rough resemblance with the well-understood behavior of the coherent states of the harmonic oscillator cannot be dismissed out of hand.
References
- [1] Amiot, E.: Music Through Fourier Space, Discrete Fourier Transform in Music Theory. Springer, Cham (2016)
- [2] Arthur, C.: When the Leading Tone Doesn’t Lead: Musical Qualia in Context. Ph.D. thesis, Ohio State University (2016),
- [3] Blutner, R., beim Graben, P.: Gauge models of musical forces. Journal of Mathematics and Music 15(1), 17 – 36 (2021)
- [4] Deutsch, D., Feroe, J.: The internal representation of pitch sequences in tonal music. Psychological Review 88(6), 503 – 522 (1981).
- [5] beim Graben, P.: Gauge symmetries of musical and visual forces. Journal of Mathematics and the Arts 17(3-4), 347 – 382 (2023)
- [6] beim Graben, P., Blutner, R.: Quantum approaches to music cognition. Journal of Mathematical Psychology 91, 38 – 50 (2019)
- [7] beim Graben, P., Mannone, M.: Musical pitch quantization as an eigenvalue problem. Journal of Mathematics and Music 14(3), 329 – 346 (2020)
- [8] beim Graben, P., Noll, T.: Unified musical quantum models for tonal attraction (in prep)
- [9] Huron, D.: Sweet Anticipation: Music and the Psychology of Expectation. MIT Press, Cambridge (MA) (2006)
- [10] Krumhansl, C.L., Kessler, E.J.: Tracing the dynamic changes in perceived tonal organization in a spatial representation of musical keys. Psychological Review 89(4), 334 – 368 (1982)
- [11] Lerdahl, F.: Tonal Pitch Space. Oxford University Press, NewYork (2001)
- [12] McWeeny, R., Coulson, C.A.: Quantum mechanics of the anharmonic oscillator 44(3), 413 – 422 (1948)
- [13] von Neumann, J.: https://de.wikipedia.org/wiki/John_von_Neumann
- [14] Noll, T., beim Graben, P.: Quantum-musical explorations on . In: Montiel, M., Agustín-Aquino, O.A., Gómez, F., Kastine, J., Lluis-Puebla, E., Milam, B. (eds.) Mathematics and Computation in Music. pp. 369 – 375. Springer, Cham (2022)
- [15] Regener, E.: Pitch Notation and Equal Temperament: A Formal Study, Occasional Papers, vol. 6. University of California Press, Berkeley (1973)
- [16] Russell, S., Norvig, P.: Artificial Intelligence: A Modern Approach. Pearson, 3rd edn. (2010)
- [17] Schrödinger, E.: Der stetige Übergang von der Mikro- zur Makromechanik. Naturwissenschaften 14(28), 664 – 666 (1926)
- [18] Temperley, D.: Music and Probability. MIT Press, Cambridge (MA) (2007)
- [19] Woolhouse, M.: Modelling tonal attraction between adjacent musical elements. Journal of New Music Research 38(4), 357 – 379 (2009)
- [20] Yust, J.: Harmonic qualities in Debussy’s ’les sons et les parfums tournent dans l’air du soir’. Journal of Mathematics and Music 11(2-3), 155–173 (2017)