Iannis Xenakis (1922–2001) arrived at composition through engineering, architecture, war, exile, and Messiaen’s classroom. In Pithoprakta, written in 1955–56, he treated individual string glissandi as lines in a statistical sound-mass. The comparison with molecules is Xenakis’s. Its limits matter as much as its audacity.


From Serial Detail to Statistical Form

Xenakis criticised the audible result of highly elaborated serial procedures: independent deterministic organisation at the note level could produce a mass whose global behaviour sounded uncontrolled. His alternative was not “randomness instead of composition.” It was control at another scale. Specify distributions, densities, and transformations for a sound-mass, then realise individual events within those constraints.

This account follows Xenakis’s framing in Formalized Music [[1]], as reported by Benoît Gibson [[2]]. It is theoretical and polemical, not a neutral census of all European music in the 1950s. Total serialism was influential, but “what everyone else was doing” would be indefensible history.


One Velocity Component, Not Three

In a classical ideal gas at equilibrium each Cartesian velocity component is Gaussian. Maxwell’s 1860 paper writes its density, for a velocity “resolved in a certain direction,” as $e^{-x^2/\alpha^2}/(\alpha\sqrt{\pi})$ and derives from it the density of the speed $v$ [[3]]:

$$f_3(v)=\frac{4}{\alpha^3\sqrt{\pi}}\,v^2e^{-v^2/\alpha^2},\qquad \alpha=\sqrt{\frac{2k_BT}{m}}. \tag{1}$$

Here $k_B$ is Boltzmann’s constant, $T$ the temperature and $m$ the molecular mass. The most probable, mean and root-mean-square speeds are $\alpha$, $2\alpha/\sqrt{\pi}$ and $\sqrt{3/2}\,\alpha$. The factor $v^2$ counts directions (a sphere of radius $v$ has area proportional to $v^2$), so the density vanishes at $v=0$.

Those equations are physical statements about a model with specified temperature, particle mass, equilibrium assumptions, and three velocity components. They are not the density Xenakis printed for Pithoprakta, which Gibson quotes from Formalized Music [[2]]:

$$f(v)=\frac{2}{\alpha\sqrt{\pi}}e^{-v^2/\alpha^2},\qquad v\ge0. \tag{2}$$

Here $v$ is the speed of a glissando, its change of pitch per unit time, whatever its direction. Eq. (2) has no factor $v^2$. It is Maxwell’s law for a single velocity component folded onto $v\ge0$, a half-Gaussian: the factor 2 makes it integrate to one over positive speeds, and Xenakis’s probability tables are differences of $\mathrm{erf}(v/\alpha)$ [[4]], [[5]]. Its most probable value is zero, its mean $\alpha/\sqrt{\pi}$ and its root-mean-square $\alpha/\sqrt{2}$. Du Toit reports Xenakis writing $\alpha=\sqrt{2}\,s$ for a standard deviation $s$, consistent with Eq. (2); only Xenakis’s other gloss, $\alpha$ as the “quadratic mean” of the speeds, is loose by that factor [[4]], [[5]].

In Musiques formelles Xenakis treats the glissando as a speed, “une représentation vectorielle à une dimension” whose magnitude is the hypotenuse of a right triangle built on the melodic interval; Barthel-Calvet, who quotes the passage, adds that he applies the law of Maxwell and Boltzmann to these speeds [[6]]. Reading Eq. (2) as the one-component law is my inference from its form, not a choice the sources record. At the same $\alpha$ (for a gas, the same temperature) the one-component law peaks at zero and has exactly half the mean of the three-dimensional one; 52 per cent of its speeds lie below $\alpha/2$, against 8 per cent of molecular speeds (Fig. 1).

Figure 1. Without Maxwell’s factor $v^2$, the most probable glissando speed is zero. Solid: Eq. (2); dashed: Eq. (1); same $\alpha$, both multiplied by $\alpha$; dotted lines at the means.

What Pithoprakta Computes

The often-analysed passage is bars 52–60 in Xenakis’s description (52–59 in Gibson’s and Solomos’s), about 18.5 seconds of music. In Formalized Music Xenakis writes that “a total of 1148 speeds, distributed in 58 distinct values according to Gauss’ law, have been calculated and traced for this passage” [[2]]. By my arithmetic, 1148 over 18.5 s is about 62 lines a second. Even the total is reported differently across records: 1142 in his 1956 Gravesaner Blätter article, 1148 in Musiques formelles (1963) and its English version, and 1146 in Makis Solomos’s own count [[7]]. That discrepancy is better evidence than a falsely exact number.

The string parts trace glissando segments with those slopes, each set off by a pizzicato. The resulting geometry gives the listener a controlled statistical mass without requiring a single melody to organise every line. The distribution constrains the material; Xenakis still chose its scale, broke lines to move them between registers, thinned some registers while thickening others [[2]], and made compositional decisions not generated by the probability density.

In an earlier version of this post I called each of the work’s 46 strings [[8]] a molecule and promised that a histogram of the score would reproduce Maxwell’s speed law. It would not: the law Xenakis computed from is Eq. (2), not Eq. (1). “Heating” has a definite counterpart in his model, raising $\alpha$, which makes steep glissandi more common. No Boltzmann constant, molecular mass, or thermometer turns the score into a gas.


Poisson Density Belongs to Achorripsis

For independent events with mean count $\lambda T$ in a window of duration $T$, the Poisson model is

$$P(N=k)=\frac{(\lambda T)^k e^{-\lambda T}}{k!},$$

with mean and variance $\lambda T$.

Xenakis’s canonical use of Poisson allocation is Achorripsis (1956–57), for 21 instruments [[8]]. Its matrix has seven rows, one per timbre class, and 28 columns of about 15 seconds each [[4]], [[9]], and Xenakis set $\lambda T=0.6$ events per cell [[2]], [[4]]. Over $7\times28=196$ cells that predicts

$$196\,P(N=k)\approx107.6,\ 64.5,\ 19.4,\ 3.9,\ 0.6\qquad\text{for }k=0,\dots,4,$$

and the matrix has 107 empty cells and 65, 19, 4 and 1 cells holding one, two, three and four events [[4]], [[9]]: 119 events, 0.61 per cell. An event is a passage for one timbre class at an average of about five sounds per bar [[9]]. In the glissando cells Xenakis made $\alpha$ in Eq. (2) proportional to the cell’s density [[5]]: a denser cell is, in the analogy, a hotter one.

The matrix leaves room for composer choice; it is not a mechanical printout of random numbers. It assigns no events to sections 8 and 24, yet the score gives both passages of minimal density, “presumably so that the work would not sound as if it had come to a premature conclusion,” and the closing tremolo glissando gesture is “not accounted for in the probability matrix” [[9]]. Edward Childs found Xenakis “most rigorous” with the exponential law for the time between events and less so with pitch intervals and glissando speeds [[4]].

Assigning each law to its work changes the historical argument. Xenakis developed related stochastic techniques across works. Combining them into one imaginary thermodynamic control panel makes the story tidier than the scores.


Markov Processes in Analogique A and B

A transition matrix satisfies

$$P_{ij}\ge0,\qquad \sum_jP_{ij}=1,$$

and a stationary distribution $\mu$, when it exists, satisfies $\mu P=\mu$. For a finite irreducible aperiodic chain, the stationary distribution is unique and the chain converges to it.

Analogique A (1958), for nine strings, uses statistically selected ranges of frequency, intensity, and density whose changes follow transition probabilities. The Xenakis catalogue describes eight “sonic states” chained by a more general Markovian process that tends toward a stationary state, is perturbed, and returns to equilibrium [[8]]. Analogique B (1959), for tape, shares the structure while adding a granular conception of sound [[8]]. For Markov chains as a compositional model more generally, see Ames’s survey [[10]].

It is tempting to say that Xenakis selected a spectral gap as a compositional memory knob. The general mathematics supports such an analysis of a specified matrix, but the cited historical sources do not show him framing the works in those modern terms.


Game Structures in Duel and Stratégie

For a two-player zero-sum matrix game, mixed strategies $x$ and $y$ give expected payoff

$$E(x,y)=x^TAy.$$

Von Neumann’s minimax theorem [[11]] equates the maximum guaranteed payoff and minimum upper bound under the theorem’s finite-game assumptions. This is the formal background for Xenakis’s two-orchestra game works.

In Duel (1959) both orchestras draw on the same six types of sonic event, one of them silence [[12]]; pairing them gives, by my count, a 6-by-6 matrix. Stratégie (1962) drops silence and adds thirteen combinations of its six basic tactics, for 19 tactics and a 19-by-19 matrix [[12]]. Conductors choose musical tactics while a scoring scheme evaluates combinations. The structure creates contingent performance paths.

Minimax mathematics does not guarantee dramatic tension, prove that no tactic dominates without inspecting the matrix, or show that conductors compute an equilibrium on the podium. The game is a compositional and performance protocol, not evidence that every performance realises optimal play.


What the Physics Contributes

The strongest version of the analogy is structural. Kinetic theory offers a way to think about a distribution of many elements rather than a privileged trajectory. Xenakis maps that way of thinking into pitch–time geometry. The mapping is neither a thermodynamic experiment nor decorative vocabulary: it is a procedure that constrains a score.

That middle position is enough. It preserves the composer’s calculation without claiming that the orchestra has pressure, temperature, collisions, or a measured Maxwell–Boltzmann speed histogram. It also preserves the composer: distributions do not choose their own parameters, instrumentation, placement, or form.

When I hear the glissando mass, I hear statistical organisation made audible. The orchestra is not a gas. The calculation is still in the score.

Literature and work-catalogue record checked through 2026-10-24. I could not open Formalized Music, Musiques formelles or the Gravesaner Blätter article; everything quoted from them is second-hand, through the sources cited.


References

[1] Xenakis, I. (1992). Formalized Music: Thought and Mathematics in Composition (rev. ed., additional material compiled and edited by S. Kanach). Pendragon Press. French original: Musiques formelles, La Revue musicale 253–254 (1963).

[2] Gibson, B. (2023). Stochastic music. In D. Exarchos (Ed.), A Xenakis Dictionary. Association Les Amis de Xenakis. https://www.iannis-xenakis.org/en/stochastic-music/

[3] Maxwell, J. C. (1860). Illustrations of the dynamical theory of gases. Part I. On the motions and collisions of perfectly elastic spheres. Philosophical Magazine, 4th series, 19(124), 19–32. DOI: 10.1080/14786446008642818. Reprinted with Parts II and III in The Scientific Papers of James Clerk Maxwell, vol. 1 (1890), 377–409; Prop. IV on pp. 380–381. https://archive.org/details/scientificpapers01maxwuoft

[4] Childs, E. (2002). Achorripsis: A sonification of probability distributions. In Proceedings of the 2002 International Conference on Auditory Display (Kyoto). https://hdl.handle.net/1853/51332

[5] du Toit, P. J. (2009). Iannis Xenakis (1922–2001): An examination of the implementation of stochastic procedures in selected compositions [MMus thesis, Stellenbosch University]. https://scholar.sun.ac.za/handle/10019.1/1645

[6] Barthel-Calvet, A.-S. (2006). La vitesse, mesure de la continuité sonore chez Xenakis. In M. Solomos, A. Georgaki & G. Zervos (Eds.), Definitive Proceedings of the International Symposium Iannis Xenakis (Athens, May 2005). First printed in the symposium’s conference proceedings (2005), 177–191. https://cicm.univ-paris8.fr/ColloqueXenakis/papers/Barthel-Calvet.pdf

[7] Solomos, M. (2023). Sonority (J. Higgins, Trans.). In D. Exarchos (Ed.), A Xenakis Dictionary. Association Les Amis de Xenakis. https://www.iannis-xenakis.org/en/sonority/

[8] Les Amis de Xenakis. Work catalogue entries, checked 24 October 2026: Pithoprakta (strings 12.12.8.8.6), Achorripsis, Analogique A and Analogique B.

[9] Squibbs, R. (2006). The composer’s flair: Achorripsis as music. In M. Solomos, A. Georgaki & G. Zervos (Eds.), Definitive Proceedings of the International Symposium Iannis Xenakis (Athens, May 2005). First printed in A. Georgaki & M. Solomos (Eds.), International Symposium Iannis Xenakis: Conference Proceedings (2005), 258–264. https://cicm.univ-paris8.fr/ColloqueXenakis/papers/Squibbs.pdf

[10] Ames, C. (1989). The Markov process as a compositional model: A survey and tutorial. Leonardo, 22(2), 175–187. DOI: 10.2307/1575226

[11] von Neumann, J. (1928). Zur Theorie der Gesellschaftsspiele. Mathematische Annalen, 100, 295–320. DOI: 10.1007/BF01448847

[12] Harley, J. (2023). Game theory. In D. Exarchos (Ed.), A Xenakis Dictionary. Association Les Amis de Xenakis. https://www.iannis-xenakis.org/en/game-theory/


Changelog

  • 2026-01-14: Corrected the tactic/matrix counts for Duel and Stratégie, the 1963 premiere date, and the account of Xenakis approaching Honegger.
  • 2026-07-11: Corrected the Pithoprakta distribution and count uncertainty, moved the Poisson example to Achorripsis, bounded the Markov and game-theory claims, and removed literal thermodynamic and historical overclaims.
  • 2026-10-24: Corrected the glissando law: Eq. (2) is a half-Gaussian, Maxwell’s law for one velocity component folded onto positive speeds (mean $\alpha/\sqrt{\pi}$), not a “one-dimensional Gaussian” separate from Maxwell’s law. Wrote Maxwell’s speed law in the same $\alpha=\sqrt{2k_BT/m}$ and added Fig. 1. “Heating”, previously a perceptual analogy pending a specified parameter, now names it: raising $\alpha$, which Xenakis made proportional to cell density in Achorripsis. Attributed the formula to Gibson quoting Formalized Music (not the work catalogue), 1142 to the 1956 Gravesaner Blätter article and 1146 to Solomos’s own count. Added the Achorripsis matrix numbers and documented departures from the matrix. Removed the Meta-Xenakis entry, which nothing cited and whose editor, year and DOI were wrong; completed the Squibbs entry.