In Schläfli notation — the system for naming regular star polygons — these are $\{7/2\}$ and $\{7/3\}$ [1]. Both appear in Tool’s artwork, in Thelemic symbolism, in medieval Islamic geometric patterns, and on the floor plans of cathedrals. They are the most visually intricate star polygons that can be drawn in a single closed stroke before the figure becomes illegible.

Both of them have a property that five-pointed and six-pointed stars do not share: they visit every vertex before closing. This is a consequence of 7 being prime. And it turns out to matter for how rhythmic accent cycles are built.

The Schläfli Symbol

A regular star polygon $\{n/k\}$ is constructed by placing $n$ points evenly on a circle and connecting every $k$-th point in sequence until the path closes. The structural key is a single number:

If $d = 1$, the traversal visits all $n$ vertices before returning to the start — a single connected figure. If $d > 1$, the path visits only $n/d$ vertices before closing, and the full figure consists of $d$ separate copies of the smaller star $\{(n/d)\,/\,(k/d)\}$.

The most familiar example of the disconnected case: $\{6/2\}$, the Star of David. Here $\gcd(6,2) = 2$, so the figure breaks into two copies of $\{3/1\} = \{3\}$ — two overlapping equilateral triangles. The traversal starting at vertex 1 visits $1 \to 3 \to 5 \to 1$, leaving vertices 2, 4, 6 entirely unvisited.

The pentagram $\{5/2\}$ is connected: $\gcd(5,2)=1$, traversal $1 \to 3 \to 5 \to 2 \to 4 \to 1$, all five vertices.

For $n=7$:

• $\{7/2\}$: $\gcd(7,2)=1$, traversal $1 \to 3 \to 5 \to 7 \to 2 \to 4 \to 6 \to 1$, all seven vertices.
• $\{7/3\}$: $\gcd(7,3)=1$, traversal $1 \to 4 \to 7 \to 3 \to 6 \to 2 \to 5 \to 1$, all seven vertices.

Both connected. Neither leaves any vertex unvisited.

The Group Theory

The traversal of $\{n/k\}$ is an instance of a standard construction in modular arithmetic: the orbit of an element under repeated addition in $\mathbb{Z}/n\mathbb{Z}$.

Label the $n$ vertices $0, 1, \ldots, n-1$. Starting at vertex 0, the traversal visits:

The orbit of 0 under the action of $+k$ is the subgroup of $\mathbb{Z}/n\mathbb{Z}$ generated by $k$. By a standard result, this subgroup has size $n / \gcd(n,k)$.

• When $\gcd(n,k) = 1$: orbit size $= n$. The traversal visits every vertex.
• When $\gcd(n,k) = d > 1$: orbit size $= n/d$. The traversal visits only a fraction of the vertices.

For prime $n$: $\gcd(n,k) = 1$ for every $1 \leq k \leq n-1$, without exception. Every traversal is complete. There is no step size that traps the path in a proper sub-orbit before visiting all vertices. This follows directly from the fact that a prime has no divisors other than 1 and itself, so $\mathbb{Z}/p\mathbb{Z}$ has no non-trivial subgroups (Lagrange’s theorem: any subgroup of a group of prime order must have order 1 or $p$).

This is the specific property that makes 7 — and any prime — rhythmically fertile.

The Contrast with Six

The comparison with $n = 6$ is the clearest illustration.

In $\mathbb{Z}/6\mathbb{Z}$, the possible step sizes are 1, 2, 3, 4, 5. Their orbits:

The only step sizes that visit all six vertices are 1 and 5 — both of which just traverse the hexagon itself, not a star. Every non-trivial star polygon on six points gets trapped. $\{6/2\}$ visits only half the vertices. $\{6/3\}$ visits only two. There is no connected six-pointed star that isn’t either the hexagon or a compound figure.

In $\mathbb{Z}/7\mathbb{Z}$, every step from 2 to 5 generates the full group:

All four non-trivial step sizes give connected traversals. Both are stars. Both visit every vertex. This is not a coincidence: it is the algebraic signature of primality.

From Geometry to Rhythm

The connection to drumming is direct. Here is the mechanism.

Consider a repeating rhythmic figure of 7 beats — a bar of 7/8, say, with positions 1 through 7. An earlier post discussed Euclidean rhythms: the algorithm that distributes $k$ onset positions as evenly as possible among $n$ slots. That is a problem of selection — which of the $n$ positions to activate.

The star polygon traversal asks a different question. Given that all $n$ positions are present, in what order of emphasis should they be related, such that each accent is a fixed distance from the last? The traversal of $\{n/k\}$ answers this: accent position $1$, then $1+k$, then $1+2k$, and so on modulo $n$.

For $\{7/2\}$: the accent cycle within a single bar runs $1 \to 3 \to 5 \to 7 \to 2 \to 4 \to 6$. Each featured beat is two positions ahead of the last.

Now project this across multiple bars. In bar 1, the primary accent sits on beat 1. In bar 2, if the accent shifts by 2, it lands on beat 3. Bar 3: beat 5. Bar 4: beat 7. Bar 5: beat 2. Bar 6: beat 4. Bar 7: beat 6. Bar 8: beat 1 again.

The accent takes seven bars to return to its starting position. Because $\gcd(2,7) = 1$, the step of 2 generates all of $\mathbb{Z}/7\mathbb{Z}$: every beat position receives the accent exactly once before the cycle resets. The resulting large-scale figure is $7 \times 7 = 49$ beats long — a super-phrase built from a single local rule.

The $\{7/3\}$ traversal generates the same exhaustiveness with a different path. Step 3 gives $1 \to 4 \to 7 \to 3 \to 6 \to 2 \to 5$: a seven-bar accent cycle that visits every position before repeating, but with wider spacing between accented beats, creating a different feel over the same underlying meter.

A six-beat figure with step 2 cannot do this. The accent visits only beats 1, 3, 5 — half the cycle — and loops back without touching beats 2, 4, 6. A drummer building phrase-level architecture from a six-beat grid is working with a more fragmented material.

Two Problems, One Prime

It is worth stating the relationship between the star polygon approach and Euclidean rhythms precisely, because the two are sometimes conflated [2].

The Euclidean algorithm distributes $k$ onsets among $n$ positions with maximal evenness. The result is a subset of the $n$ positions — a selection. The primality of $n$ matters here too: because $\gcd(k,p) = 1$ for prime $p$ and any $1 \leq k \leq p-1$, the Euclidean rhythm $E(k,p)$ always achieves its theoretical maximum of evenness. There are no divisibility shortcuts that cause clumping.

The star polygon traversal selects no subset — it relates all $n$ positions via a cyclic permutation. The primality of $n$ matters here because it guarantees that every non-trivial cyclic permutation (every step size $k$ with $1 < k < n$) generates the full group, visiting all positions before repeating.

Same arithmetic property — $\gcd(k,p) = 1$ for all non-zero $k$ — but the two problems ask different things of it. Euclidean rhythms use it to guarantee dense coverage. Star polygon traversals use it to guarantee no sub-orbit trapping.

The Compound Structure

Written out explicitly, the $\{7/2\}$ accent pattern over seven bars looks like this — with bold marking the featured beat in each bar:

Each bar is metrically identical. The large-scale accent — which beat carries the phrase-level emphasis — traces the traversal path of the $\{7/2\}$ star polygon across the seven-bar cycle.

This is the kind of large-scale rhythmic architecture visible in a great deal of Tool’s output. Whether Danny Carey explicitly constructs accent cycles from star polygon traversal paths, or whether the same structure emerges from his intuitive sense of how prime time signatures behave, produces the same result. The mathematics and the musical instinct point toward the same pattern.

Why the Heptagram

The full mathematical picture of why seven-fold symmetry is special — why the regular heptagon cannot be constructed by compass and straightedge, what the minimal polynomial of $\cos(2\pi/7)$ implies about the heptagon’s position outside the constructible world, and how the Galois group of the cyclotomic field over $\mathbb{Q}$ carries the obstruction — is developed in the companion post The Impossible Heptagon.

The short version, for the purposes of this post: seven is the smallest odd prime that is not a Fermat prime ($2^{2^j}+1$). This algebraic accident places it outside the reach of ruler-and-compass construction — the heptagon exists as an ideal but cannot be manifested by the classical tools. Its star polygons are the accessible shadows of an inaccessible form. And its primality, in both the constructibility sense and the traversal sense, is precisely what makes it inexhaustible as a rhythmic resource.

The Fibonacci structure in “Lateralus” [3], the group theory underlying twelve-tone equal temperament [4], and the Euclidean rhythm algorithm [5] are all different facets of the same observation: mathematical structure, introduced as compositional constraint, generates musical complexity that cannot easily be produced by intuition alone. The star polygon is another instance. The drummer who keeps a heptagram on his kit has found, by a non-mathematical route, an object with a precise and interesting mathematical identity.

References

[1] Coxeter, H.S.M. (1973). Regular Polytopes (3rd ed.). Dover. Ch. 2.

[2] Toussaint, G. (2013). The Geometry of Musical Rhythm: What Makes a “Good” Rhythm Good? CRC Press.

[3] See Fibonacci and Lateralus on this blog.

[4] See Twelve-TET and Group Theory on this blog.

[5] See Euclidean Rhythms on this blog.

On 12 October 2024, a retired NVIDIA engineer named Luke Durant announced that he had found the 52nd known Mersenne prime. The number is $2^{136{,}279{,}841} - 1$, and writing it out in decimal requires 41,024,320 digits. Durant had organised a cloud network of GPU servers spread across 17 countries — essentially repurposing the hardware that normally trains language models to instead do modular arithmetic on numbers with tens of millions of digits. The verification alone took about 51 days of computation.

This is the kind of thing that makes headlines, and it deserves them. Mersenne primes are rare and verifying them is genuinely hard. But if I am honest, the more interesting prime story of the last half-century is not about the record-breaking number. It is about a conversation over tea in Princeton in 1972, and the increasingly hard-to-dismiss suspicion that the prime numbers are, in a precise statistical sense, quantum energy levels.

When I say “quantum energy levels,” I mean it almost literally — not as a metaphor. Let me explain.

The Riemann Zeta Function Encodes the Primes

Start with the most famous function in number theory. For $\operatorname{Re}(s) > 1$, the Riemann zeta function is defined by the series

This converges nicely and defines an analytic function. But the real reason to care about it is Euler’s product formula:

This is not obvious — it follows from unique prime factorisation, essentially — but its implications are enormous. The product runs over all primes, and each prime contributes a factor. The primes are encoded in the analytic structure of $\zeta$. If you know $\zeta$, you know the primes; if you understand the zeros of $\zeta$, you understand their distribution.

Riemann’s 1859 paper made this explicit (Riemann, 1859). He showed that $\zeta$ extends analytically to the whole complex plane (minus a simple pole at $s = 1$), and he wrote down an explicit formula connecting the prime-counting function

to the zeros of $\zeta$. The formula is

where $\operatorname{Li}(x) = \int_2^x \frac{dt}{\ln t}$ is the logarithmic integral and the sum runs over the non-trivial zeros $\rho$ of $\zeta$.

What are the non-trivial zeros? The zeta function has trivial zeros at the negative even integers $-2, -4, -6, \ldots$ — boring, understood. The non-trivial zeros lie in the critical strip $0 < \operatorname{Re}(s) < 1$, and their imaginary parts are what drive the oscillatory corrections to $\pi(x)$. Each zero $\rho = \frac{1}{2} + it_n$ contributes a term that oscillates like $x^{1/2} \cos(t_n \ln x)$. The prime distribution is a superposition of these oscillations, one per zero.

The Riemann Hypothesis is the claim that all non-trivial zeros lie on the critical line $\operatorname{Re}(s) = \frac{1}{2}$. It has been verified numerically for the first $10^{13}$ zeros (Gourdon, 2004; building on earlier high-height computations by Odlyzko, 1987). It has not been proved. It remains, after 165 years, the most important unsolved problem in mathematics.

Tea with Dyson

In 1972, Hugh Montgomery was visiting the Institute for Advanced Study in Princeton. He was working on a specific question: if you take the imaginary parts of the non-trivial zeros of $\zeta$ and normalise them so that their mean spacing is 1, what is the distribution of spacings between them?

More precisely, he was computing the pair correlation function of the normalised zeros. If $\tilde{\gamma}_n$ are the normalised imaginary parts (ordered $\tilde{\gamma}_1 \leq \tilde{\gamma}_2 \leq \cdots$), the pair correlation function $R_2(r)$ measures the density of pairs $(\tilde{\gamma}_m, \tilde{\gamma}_n)$ with $\tilde{\gamma}_n - \tilde{\gamma}_m \approx r$.

Montgomery found — subject to certain assumptions about the behaviour of $\zeta$ — that

(Montgomery, 1973)

He mentioned this to Freeman Dyson over tea. Dyson — who had spent years on quantum mechanics and random matrix theory — recognised the formula immediately. That expression, $1 - (\sin \pi r / \pi r)^2$, is exactly the pair correlation function of eigenvalues of random matrices drawn from the Gaussian Unitary Ensemble.

Montgomery had not been thinking about quantum mechanics. Dyson had not been thinking about primes. The formula matched.

The Gaussian Unitary Ensemble

Let me say a few words about where that formula comes from in physics, because it is not obvious.

The Gaussian Unitary Ensemble (GUE) is a probability distribution over $N \times N$ Hermitian matrices. Specifically, it is the distribution proportional to $e^{-\operatorname{tr}(H^2)}$ on the space of Hermitian matrices, which is invariant under conjugation $H \mapsto U H U^\dagger$ for any unitary $U$. The entries on the diagonal are real Gaussians; the off-diagonal entries are complex Gaussians with independent real and imaginary parts.

In the limit $N \to \infty$, the eigenvalues of a GUE matrix distribute globally according to Wigner’s semicircle law. But the local statistics — the fine-grained distribution of spacings between nearby eigenvalues — follow a universal law. The pair correlation function is

This distribution has a crucial qualitative feature called level repulsion: as $r \to 0$, $R_2(r) \to 0$. Eigenvalues of random Hermitian matrices strongly avoid each other. A Poisson distribution — which is what you would get for eigenvalues that were statistically independent — would give $R_2(r) = 1$ everywhere, with no such repulsion. The GUE formula suppresses small gaps quadratically: $R_2(r) \sim \pi^2 r^2 / 3$ for small $r$.

Why does GUE statistics arise in physics? This is the content of the Bohigas-Giannoni-Schmit conjecture (1984), which by now has overwhelming numerical support: quantum systems whose classical limit is chaotic and which lack time-reversal symmetry have energy level statistics described by the GUE. Systems with time-reversal symmetry fall into the Gaussian Orthogonal Ensemble (GOE), which has a different but related formula. Nuclear energy levels, quantum billiards with the right shapes, molecular spectra — all of them, when appropriately normalised, show GUE or GOE statistics.

The universality is the point. It does not matter what the specific Hamiltonian is. If the system is sufficiently chaotic, the eigenvalue statistics are universal.

Odlyzko’s Computation

Montgomery’s result was conditional and covered only a limited range of $r$. The natural next step was numerical verification: actually compute a large number of Riemann zeros and measure their pair correlation.

Andrew Odlyzko did exactly this, in a series of computations beginning in the 1980s. The results were striking (Odlyzko, 1987). He computed millions of zeros with high precision and compared their empirical pair correlation to the GUE prediction. The agreement was not merely qualitative — it was quantitatively exact, to within the statistical error of the sample.

Odlyzko then pushed further. He computed zeros near the $10^{20}$-th zero, far out on the critical line. Same statistics. He computed zeros near the $10^{22}$-th zero. Same statistics. The agreement held regardless of how far up the critical line one went. This is not a small-sample artifact and it is not coincidence, or at least it would be an extraordinary coincidence of a kind that mathematics has never before encountered.

The plots from Odlyzko’s computations are, in my view, some of the most beautiful images in mathematics. You draw the GUE prediction — a smooth curve, starting at zero, rising to approach 1 — and you overlay the empirical histogram from the Riemann zeros. They are the same curve.

Berry, Keating, and the Missing Hamiltonian

If the zeros of $\zeta$ are energy levels, there should be a Hamiltonian $H$ — a self-adjoint operator — whose spectrum is exactly $\{t_n\}$, the imaginary parts of the non-trivial zeros (assuming the Riemann Hypothesis, so that all zeros are of the form $\frac{1}{2} + it_n$).

In 1999, Michael Berry and Jon Keating proposed a candidate (Berry & Keating, 1999). Their suggestion was the classical Hamiltonian

where $x$ is position and $p$ is momentum, quantized with appropriate symmetrization:

Classically, $H = xp$ describes a system in which the phase-space trajectories are hyperbolas $xp = E = \text{const}$, and the motion is $x(t) = x_0 e^t$, $p(t) = p_0 e^{-t}$ — exponential expansion in position, contraction in momentum. This is essentially the dynamics of an unstable fixed point, and it is classically chaotic in the appropriate sense.

The semiclassical (WKB) approximation gives an eigenvalue counting function

which matches Riemann’s formula for the number of zeros of $\zeta$ with imaginary part up to $T$:

This is not a coincidence: the correspondence is exact at the level of the smooth counting function. The hard part is the oscillatory corrections — and those require the specific eigenvalues, which requires knowing the boundary conditions.

The problem is that $\hat{H} = \frac{1}{2}(\hat{x}\hat{p} + \hat{p}\hat{x})$ as an operator on $L^2(\mathbb{R})$ is not bounded below and has a continuous spectrum, not a discrete one. Turning it into an operator with a discrete spectrum matching the Riemann zeros requires boundary conditions that have not been found. This is the crux: Berry and Keating have the right classical system, but the quantum boundary conditions are missing.

What would be profound about finding $\hat{H}$? If $\hat{H}$ is self-adjoint and bounded below ($\hat{H} \geq 0$), its eigenvalues are all non-negative real numbers. If those eigenvalues are the imaginary parts of the zeros, then all zeros have real part exactly $\frac{1}{2}$ — which is the Riemann Hypothesis. A proof of the existence of such a Hamiltonian would, in one stroke, resolve the most important open problem in mathematics.

Primes as Periodic Orbits: The Gutzwiller Analogy

The quantum chaos connection goes deeper than pair correlations. In semiclassical quantum mechanics, the Gutzwiller trace formula relates the density of quantum energy levels to a sum over classical periodic orbits:

where the sum runs over all classical periodic orbits $\gamma$, $S_\gamma$ is the classical action of the orbit, $A_\gamma$ is an amplitude, and $\phi_\gamma$ is a phase (Maslov index correction). The smooth part $\bar{d}(E)$ comes from the Thomas-Fermi approximation; the oscillatory part encodes quantum interference between orbits.

The direct analogue in number theory is the explicit formula for the prime-counting function. Written as a formula for the oscillatory part of the zero-counting function, it reads

where $\psi(x) = \sum_{p^k \leq x} \ln p$ is the Chebyshev function and the sum is over non-trivial zeros $\rho$.

Comparing these two formulas term by term: the zeros $\rho$ of $\zeta$ play the role of the quantum energy levels $E_n$; the primes $p$ — and their prime powers $p^k$ — play the role of the classical periodic orbits $\gamma$. The “action” of the orbit corresponding to $p^k$ is $k \ln p$. The primes are the primitive periodic orbits; $p^k$ is the $k$-th traversal of that orbit.

This is not a metaphor or a loose analogy. The Selberg trace formula — developed for the Laplacian on hyperbolic surfaces — makes this correspondence rigorous in a related setting: the periodic geodesics on a hyperbolic surface play the role of primes, and the eigenvalues of the Laplacian play the role of Riemann zeros (Rudnick & Sarnak, 1996). The Riemann zeta function is the limit of a family of such systems, in some sense that is still being made precise.

I find it remarkable that the logarithms of primes — the most elementary sequence in arithmetic — appear as lengths of orbits in what would be a quantum chaotic system. Each prime contributes an oscillation to $\psi(x)$ with “frequency” proportional to its logarithm. You are, in a sense, hearing the primes as quantum interference.

This connects to a theme that comes up elsewhere on this blog. The falling cat problem involves Berry phase and geometric holonomy — again a situation where deep structure emerges from symmetry and topology. The Schrödinger cat in quantum computing involves the spectacular fragility of quantum coherence. The Riemann zeros are, if the conjecture is right, a quantum system that has never decohered — a perfectly coherent spectrum hiding inside the most ancient problem in mathematics.

A Brief Detour: Maynard and Primes Without Digits

While we are talking about primes, I cannot resist a detour through two results of James Maynard, who received the Fields Medal in 2022.

The first concerns bounded gaps. Euclid proved that there are infinitely many primes. The Twin Prime Conjecture says there are infinitely many pairs of primes $(p, p+2)$. This remains open. But in 2013, Yitang Zhang proved something extraordinary: there are infinitely many pairs of primes differing by at most 70,000,000 (Zhang, 2014). The bound is large, but the qualitative statement — that gaps between primes are bounded infinitely often — was completely new. Shortly thereafter, Maynard independently proved a much stronger result using the Maynard-Tao sieve: infinitely many prime pairs with gap at most 600 (Maynard, 2015). A crowdsourced effort (Polymath8b) brought the bound down to 246. The Twin Prime Conjecture remains open, but 246 is a long way from 70,000,000.

The second result is stranger. Maynard proved in 2016 that for any decimal digit $d \in \{0, 1, \ldots, 9\}$, there are infinitely many primes whose decimal representation contains no instance of $d$. There are infinitely many primes with no $7$ in their decimal expansion. There are infinitely many primes with no $3$. The proof uses techniques from analytic number theory, specifically exponential sum estimates and sieve methods, and the result holds not just for base 10 but for any base.

This is one of those results that sounds impossible on first hearing. Surely removing an entire digit should make most large numbers unavailable, so the primes run out? Not so. The density of such “digitless” numbers thins out, but not fast enough to eliminate infinitely many primes.

The 52nd Mersenne Prime and What We Do Not Know

Return to $M_{136{,}279{,}841} = 2^{136{,}279{,}841} - 1$. Mersenne primes have the form $2^p - 1$ where $p$ is a prime (though not all such numbers are prime — $2^{11} - 1 = 2047 = 23 \times 89$). They are tested via the Lucas-Lehmer primality test: define the sequence

Then $M_p = 2^p - 1$ is prime if and only if $s_{p-2} \equiv 0 \pmod{M_p}$.

The test requires $p - 2$ squarings modulo $M_p$. Each squaring involves numbers with roughly $p$ digits, and modular reduction modulo $M_p = 2^p - 1$ is cheap because it reduces to bit-shifts. This is why GPU parallelism helps enormously: each squaring can be broken into many parallel multiplications of sub-blocks of digits. Durant’s cloud network was, in effect, a massively distributed modular arithmetic engine.

We do not know if there are infinitely many Mersenne primes. The heuristic Lenstra-Pomerance-Wagstaff conjecture says yes: the expected number of Mersenne primes $2^p - 1$ with $p \leq x$ is approximately

where $\gamma \approx 0.5772$ is the Euler-Mascheroni constant. This predicts roughly logarithmic growth in the count — consistent with the 52 known examples — but is nowhere near proved.

The known Mersenne primes do not form a sequence with obviously regular gaps. The exponents $p$ are: 2, 3, 5, 7, 13, 17, 19, 31, 61, 89, 107, 127, … and then larger, less predictable values. Whether their distribution has GUE-like statistics is not a standard research question (the sample is too small), but the question of whether the primes $p$ for which $2^p - 1$ is prime have any special structure is an active one. For now, the answer is: we do not know.

Why This Matters, and Why It Does Not Prove Anything

Let me be precise about what has and has not been established.

What has been established:

• Montgomery proved (conditionally, assuming a form of the generalised Riemann Hypothesis) that the pair correlation of Riemann zeros, for a certain range of $r$, is given by $1 - (\sin \pi r / \pi r)^2$.
• Odlyzko verified numerically — to extraordinary precision, over billions of zeros — that the full empirical pair correlation matches the GUE prediction.
• The Gutzwiller/Selberg analogy between periodic orbits and primes is mathematically precise in related settings (hyperbolic surfaces, function fields over finite fields).
• Rudnick and Sarnak proved that the $n$-point correlation functions of Riemann zeros match GUE for all $n$, subject to a plausible conjecture about $\zeta$ (Rudnick & Sarnak, 1996).

What has not been established:

• There is no known Hamiltonian $\hat{H}$ whose spectrum is the set of Riemann zeros.
• The Riemann Hypothesis remains open.
• There is no proof that the Montgomery-Odlyzko connection is anything more than an extraordinary numerical coincidence.

The broader context is the Langlands program — a still-hypothetical grand unification of number theory, algebraic geometry, and representation theory, sometimes described as a “grand unified theory of mathematics.” The Langlands correspondence predicts deep connections between $L$-functions (generalisations of $\zeta$) and representations of algebraic groups. The spectral interpretation of Riemann zeros — if it could be made precise — would fit naturally into this framework. Some researchers believe that a proof of the Riemann Hypothesis will come from the Langlands side, not from analytic number theory or quantum mechanics. Others think the quantum chaos connection is the right road. Nobody knows.

What would it mean if the connection is real? It would mean that the prime numbers — discovered by Euclid, studied for two and a half millennia, used today in every TLS handshake and RSA key — are the eigenvalues of a physical Hamiltonian. The abstract number-theoretic structure and the physical quantum mechanical structure would be not merely analogous but identical. That is a claim of the same depth as the unexpected appearance of the same partial differential equations in heat flow, diffusion, and Brownian motion: a discovery that what seemed to be different phenomena are manifestations of the same underlying law.

Or it could be a very surprising coincidence. Mathematics has a long history of producing such coincidences — the same numbers appearing in unrelated contexts for reasons that, when understood, turned out not to be coincidences at all. I suspect this is not a coincidence. But suspicion is not proof.

A Closing Reflection

I started this post with the 52nd Mersenne prime because it is the news item that prompted me to write. GPU clusters finding 41-million-digit primes are genuinely impressive technology. But I keep returning to the image of Montgomery and Dyson at tea in 1972, and the formula $1 - (\sin \pi r / \pi r)^2$ connecting two conversations that had nothing to do with each other.

I have spent some time with random matrix theory, and separately with the zeta function, and the thing that still strikes me is how clean the connection is. This is not a numerical coincidence of the form “these two quantities agree to 3 decimal places.” Odlyzko’s plots show agreement across many orders of magnitude, for zeros computed billions of entries into the sequence. The GUE curve and the empirical histogram are, visually, the same curve.

As someone trained as a physicist, I find this both encouraging and slightly unsettling. Encouraging because it suggests that the primes are not random — they have a structure, one that matches the eigenvalue repulsion of quantum chaotic systems, and that structure might be the key to proving the Riemann Hypothesis. Unsettling because it means that the quantum mechanical formalism — which I always thought was a description of a physical world — seems to be reaching into pure arithmetic, where there is no wave function, no Hilbert space, no measurement. The primes do not know they are supposed to be energy levels. And yet, statistically, they are.

If you find a flaw in this picture, or know of a result I have missed, I am genuinely interested. Peer review is welcome — open an issue on GitHub.

References

• Riemann, B. (1859). Über die Anzahl der Primzahlen unter einer gegebenen Grösse. Monatsberichte der Berliner Akademie.

• Montgomery, H. L. (1973). The pair correlation of zeros of the zeta function. Analytic Number Theory, Proc. Symp. Pure Math., 24, 181–193.

• Odlyzko, A. M. (1987). On the distribution of spacings between zeros of the zeta function. Mathematics of Computation, 48, 273–308. DOI: 10.2307/2007890

• Berry, M. V., & Keating, J. P. (1999). The Riemann zeros and eigenvalue asymptotics. SIAM Review, 41(2), 236–266. DOI: 10.1137/S0036144598347497

• Zhang, Y. (2014). Bounded gaps between primes. Annals of Mathematics, 179(3), 1121–1174. DOI: 10.4007/annals.2014.179.3.7

• Maynard, J. (2015). Small gaps between primes. Annals of Mathematics, 181(1), 383–413. DOI: 10.4007/annals.2015.181.1.7

• Rudnick, Z., & Sarnak, P. (1996). Zeros of principal L-functions and random matrix theory. Duke Mathematical Journal, 81(2), 269–322. DOI: 10.1215/S0012-7094-96-08115-6

• GIMPS (2024). 2^136279841-1 is Prime! Great Internet Mersenne Prime Search. Retrieved from https://www.mersenne.org/primes/?press=M136279841

Changelog

• 2026-02-17: Corrected the date of the Montgomery-Dyson meeting from 1973 to 1972 (the paper was published in the 1973 proceedings volume, but the meeting at the IAS took place in April 1972).

The scientific reflex here is usually impatience. “Sacred geometry” occupies an uncomfortable cultural space — mathematically dressed, factually thin, reliant on the listener not checking claims too carefully. The golden ratio does not appear everywhere in nature. Most things described as sacred in this tradition are better described as things the speaker found surprising before learning a more precise vocabulary.

But the heptagon is genuinely strange. Not for the reasons usually given. For a different reason — a theorem.

The regular heptagon cannot be constructed with compass and straightedge.

Not “it is difficult.” Not “no one has found a construction yet.” The regular seven-sided polygon — all sides equal, all interior angles equal — is provably impossible to construct using an unmarked ruler and compass in finitely many steps. This has been known since 1801.

The Classical Constraint

Greek geometry restricted its tools deliberately. An unmarked straightedge draws lines through two known points. A compass draws circles centred at a known point with a given radius. No angle trisection. No markings. No graduated instruments. Just these two operations, applied one at a time, finitely many times.

Within this constraint, a great deal is achievable. A perpendicular bisector. An equilateral triangle. A regular pentagon — which requires the golden ratio and takes some work, but is reachable. A regular hexagon (trivially: six equilateral triangles around a centre).

Then: nothing for the heptagon. Greek geometers left no construction. Medieval Islamic mathematicians, who knew the regular polygon problem well, left no construction. Albrecht Dürer, in his 1525 Underweysung der Messung, gave an approximate construction that falls short by a small but nonzero margin. Each generation encountered the same wall.

In 1796, an 18-year-old Gauss proved that the regular 17-gon is constructible — a result so unexpected that he reportedly decided at that moment to become a mathematician rather than a philologist. In his 1801 Disquisitiones Arithmeticae he gave the complete characterisation of which regular polygons are constructible and which are not [1]. The heptagon was definitively placed among the impossible.

Gauss’s Theorem

A regular $n$-gon is constructible with compass and straightedge if and only if $n$ has the form

where $k \geq 0$ and the $p_i$ are distinct Fermat primes — primes of the form $2^{2^j} + 1$.

The Fermat primes currently known:

Five Fermat primes are known, all identified by the seventeenth century. Fermat himself conjectured that all numbers of this form are prime; he was wrong from $j = 5$ onward. Whether any further Fermat primes exist remains an open problem.

The constructible regular polygons therefore include the triangle (3), square (4), pentagon (5), hexagon (6), octagon (8), decagon (10), 15-gon, 17-gon, 257-gon, 65537-gon, and products of these with powers of 2. The 65537-gon was actually fully constructed by Johann Gustav Hermes, who spent around ten years on the computation in the 1880s and deposited a manuscript reportedly filling a large trunk at the University of Göttingen, where it remains.

Seven is prime, but $7 \neq 2^{2^j} + 1$ for any $j$ — it is not a Fermat prime. Therefore the regular heptagon is not on the list. It is not constructible.

The Algebra Behind the Geometry

Why does the structure of Fermat primes determine constructibility? The connection goes through algebra [2][3].

Every compass-and-straightedge construction corresponds to solving a sequence of equations of degree at most 2. Bisecting an angle, finding an intersection of a line and a circle — each step is a quadratic operation. After $k$ such steps, the numbers reachable lie in some field extension of $\mathbb{Q}$ (the rationals) with degree over $\mathbb{Q}$ at most $2^k$. Constructibility therefore requires the degree of the relevant extension to be a power of 2.

To construct a regular $n$-gon, you need to construct the angle $2\pi/n$, which requires constructing $\cos(2\pi/n)$. The question is: over what kind of field extension does $\cos(2\pi/n)$ sit?

For $n = 7$: let $\omega = e^{2\pi i/7}$, a primitive 7th root of unity. The minimal polynomial of $\omega$ over $\mathbb{Q}$ is the 7th cyclotomic polynomial

which is irreducible over $\mathbb{Q}$, giving $[\mathbb{Q}(\omega) : \mathbb{Q}] = 6$. Since $\cos(2\pi/7) = (\omega + \omega^{-1})/2$, and since $\omega$ satisfies a degree-2 polynomial over $\mathbb{Q}(\cos 2\pi/7)$, we get

Specifically, $c = \cos(2\pi/7)$ is the root of the irreducible cubic

or equivalently, $\alpha = 2\cos(2\pi/7)$ satisfies

The three roots of this cubic are $2\cos(2\pi/7)$, $2\cos(4\pi/7)$, and $2\cos(6\pi/7)$. By Vieta’s formulas their sum is $-1$ and their product is $1$ — which can be verified directly from the identity $\cos(2\pi/7) + \cos(4\pi/7) + \cos(6\pi/7) = -1/2$.

The degree of the extension is 3. Three is not a power of 2. Therefore $\cos(2\pi/7)$ cannot be reached by any tower of quadratic extensions of $\mathbb{Q}$. Therefore the regular heptagon is not constructible. $\square$

Compare the pentagon: $\cos(2\pi/5) = (\sqrt{5}-1)/4$, satisfying the quadratic $4x^2 + 2x - 1 = 0$. Degree 2 — a power of 2. Constructible.

The 17-gon: the Galois group of $\mathbb{Q}(\zeta_{17})/\mathbb{Q}$ is $(\mathbb{Z}/17\mathbb{Z})^* \cong \mathbb{Z}/16\mathbb{Z}$, order $16 = 2^4$. The extension decomposes into four quadratic steps. This is exactly what Gauss computed at 18.

For 7: $(\mathbb{Z}/7\mathbb{Z})^* \cong \mathbb{Z}/6\mathbb{Z}$, order $6 = 2 \times 3$. The factor of 3 is the obstruction. The Galois group is not a 2-group, so the extension cannot be decomposed into quadratic steps. The heptagon is out of reach.

Sacred, Precisely

The phrase “sacred geometry” usually does work that “elegant mathematics” could do more honestly. But the heptagon is a case where something with genuine mathematical content sits underneath the mystical framing.

The Platonic tradition held that certain geometric forms exist as ideals — perfect, unchanging, more real than their physical approximations. The philosopher’s claim is that the heptagon exists in a realm beyond its material instantiation. The mathematician’s claim is: the heptagon is perfectly well-defined — seven equal sides, seven equal angles — but it cannot be reached from $\mathbb{Q}$ by the operations available to ruler and compass. You can approximate it to any desired precision. You can construct it exactly using origami, which allows angle trisection and is strictly more powerful than compass and straightedge [4]. But the classical constructive program — the one that reaches the pentagon, the hexagon, the 17-gon, the 65537-gon — cannot reach the heptagon.

There is a precise mathematical sense in which it lies outside the constructible world. Whether that constitutes sacredness is a question for a different kind of argument. But it is not nothing. The Pythagoreans were working without Galois theory; they had an intuition without the theorem. The theorem, when it came, confirmed that intuition about seven while explaining it more clearly than they could.

Carey’s intuition — that 7 sits outside the ordinary — is, by this route, formally correct.

What the Heptagram Is

The regular heptagon may be impossible to construct exactly, but the heptagram — the seven-pointed star — is perfectly drawable. Connecting every second vertex of an approximate regular heptagon gives $\{7/2\}$ in Schläfli notation [5]; connecting every third vertex gives $\{7/3\}$. Both are closed figures. Both appear throughout pre-modern symbolic traditions, which is unsurprising: they are the most intricate star polygons drawable with a single pen stroke before complexity outruns visibility.

They are also generators of rhythmic structure. Because 7 is prime, every star polygon on seven points visits all seven vertices in a single closed traversal — a property that does not hold for six-pointed or eight-pointed stars. This turns out to matter for how drum patterns are built across multiple bars. That connection — from the primality of 7 to the architecture of rhythmic accent cycles — is the subject of the companion post, Star Polygons and Drum Machines.

The broader series on mathematics in Tool’s music began with the Fibonacci structure embedded in the time signatures and syllable counts of “Lateralus” [6], and the group-theoretic structure underlying twelve-tone equal temperament provides the same algebraic scaffolding seen here [7].

References

[1] Gauss, C.F. (1801). Disquisitiones Arithmeticae. Leipzig: Fleischer. (§VII.)

[2] Stewart, I. (2004). Galois Theory (3rd ed.). CRC Press. Ch. 4.

[3] Conway, J.H. & Guy, R.K. (1996). The Book of Numbers. Springer. pp. 190–202.

[4] Hull, T. (2011). Solving cubics with creases: The work of Beloch and Lill. The American Mathematical Monthly, 118(4), 307–315. DOI: 10.4169/amer.math.monthly.118.04.307

[5] Coxeter, H.S.M. (1973). Regular Polytopes (3rd ed.). Dover. Ch. 2.

[6] See Fibonacci and Lateralus on this blog.

[7] See Twelve-TET and Group Theory on this blog.

The real answer is in number theory. Specifically, it is in the continued fraction expansion of a single irrational number: $\log_2(3/2)$. The number 12 is not a cultural choice. It is the smallest integer that gives a genuinely good rational approximation to that number — subject to the constraint that a human hand can navigate the resulting keyboard. Once you see the argument, the feeling of contingency evaporates completely. Twelve is forced on us.

Along the way, the same mathematical structure — the cyclic group $\mathbb{Z}_{12}$ — explains why Messiaen’s modes of limited transposition exist, why the circle of fifths closes exactly, and why certain chord types (augmented triads, diminished seventh chords, the whole-tone scale) have a strange self-similar quality that composers have exploited for centuries. If you want the full treatment of the Messiaen connection, I wrote a dedicated post: Messiaen, Modes, and the Group Theory of Harmony. Here I want to build the foundations from scratch, starting with the one interval that makes all of this necessary.

The interval that started everything

The perfect fifth has a frequency ratio of exactly 3:2. Play two strings in that ratio and the sound is stable, open, and unmistakably consonant — second only to the octave (2:1) in the hierarchy of simple intervals. The reason is physics: the overtone series of any vibrating string includes the fundamental frequency $f$, then $2f$, $3f$, $4f$, and so on. Two notes a perfect fifth apart share the overtone at $3f$ (for the lower note) and $2f'$ (for the upper note, where $f' = 3f/2$): those are the same frequency, $3f$. Shared overtones mean the two notes reinforce rather than fight each other. This is why the fifth sounds stable: it is literally built into the harmonic structure of physical vibration.

Humans discovered the fifth independently in ancient Greece, China, India, and Mesopotamia. It is not a cultural artifact [4]. Given that stability, it is natural to ask: can we build a complete pitch system by stacking fifths? Take a starting note, go up a fifth, up another, up another, and keep going. The notes you produce — C, G, D, A, E, B, F♯, … — are acoustically related to the starting point in a simple way, and they sound good together. This is the Pythagorean tuning system, and it underlies the construction of diatonic scales.

But here is the problem. A fifth raises the pitch by a factor of 3/2. An octave raises it by a factor of 2. These are independent: one is a power of 3 and the other a power of 2, and no power of 3/2 will ever equal a power of 2 exactly. In the language of modern mathematics, $\log_2(3/2)$ is irrational — this follows directly from the fundamental theorem of arithmetic, since no product of powers of 2 can equal a product of powers of 3. Whether it is also transcendental is an open question; a proof would follow from Schanuel’s conjecture, but that conjecture remains unresolved. What matters for tuning is the irrationality alone. Stacking pure fifths and stacking octaves are incommensurable operations. The circle of fifths can never close in pure Pythagorean tuning. We will always end up slightly sharp or flat relative to where we started.

This incommensurability is the central problem of musical tuning. Everything else — equal temperament, just intonation, meantone tuning, the Pythagorean comma, the whole apparatus of tuning theory — is a response to it.

Equal temperament and the approximation problem

In an equal temperament with $N$ notes per octave, we divide the octave into $N$ equal logarithmic steps. Each step corresponds to a frequency ratio of $2^{1/N}$. We then ask: how many steps $k$ gives the best approximation to a perfect fifth?

The condition is simply that $2^{k/N}$ should be close to $3/2$, which means $k/N$ should be close to $\log_2(3/2)$. So we need a good rational approximation to

The classical tool for finding best rational approximations is the continued fraction. Any real number $x$ can be written as

where the $a_i$ are non-negative integers (positive for $i \geq 1$), called the partial quotients. For $\log_2(3/2)$ the expansion is

The truncated continued fractions — the convergents — give the sequence of best rational approximations:

Each convergent $k/N$ corresponds to a tuning system: the denominator $N$ is the number of equal steps per octave, and the numerator $k$ is the number of steps that best approximates a fifth. So we get: 1-TET (trivial), 2-TET (trivial), 5-TET, 12-TET, 41-TET, 53-TET, 306-TET, and so on [1], [2].

The key property of convergents is that they give uniquely good approximations. No rational number with a smaller denominator comes closer to the true value than a convergent does. So 7/12 is not merely a decent approximation to $\log_2(3/2)$ — it is provably the best approximation with denominator at most 12. To do better with a denominator below 41, you cannot.

To put numbers on it: in 12-TET, the fifth is $2^{7/12} \approx 1.498307\ldots$, while the true fifth is exactly $1.500000$. The error is about 0.11%, or roughly 2 cents (hundredths of a semitone). In 53-TET, the fifth is $2^{31/53} \approx 1.499941\ldots$, an error of less than 0.004%, about 0.07 cents — essentially indistinguishable from pure. Both 12 and 53 are convergents. Intermediate values like 19-TET or 31-TET are not convergents (they are not best approximations), and their fifths, while sometimes used in experimental or microtonal music, are less accurate relative to their complexity.

Why does this matter? Because a tuning system that approximates the fifth poorly will produce harmonies that beat visibly — the slight mistuning causes the sound to waver in a way that trained ears find uncomfortable in sustained chords. A good fifth approximation is not a luxury; it is the condition for the system to be musically usable in the harmonic practice that most of the world’s music assumes.

The Pythagorean comma

Before equal temperament became standard (roughly the 18th century in Western Europe), instruments were tuned using pure Pythagorean fifths: exact 3:2 ratios, stacked on top of each other. This gives beautiful, stable individual fifths, but it collects a debt.

After stacking 12 pure fifths, you have climbed in frequency by $(3/2)^{12}$:

Meanwhile, 7 octaves is $2^7 = 128$. The ratio between these is

This is the Pythagorean comma: roughly 23.46 cents, or about a quarter of a semitone [4]. In Pythagorean tuning, the circle of fifths never closes. After 12 fifths you arrive at a note that is nominally the same pitch class as the starting point — but sharp by 23.46 cents. That final fifth, the one that “should” close the circle, sounds badly out of tune. It was historically called the “wolf fifth” because it howls.

Equal temperament solves this by distributing the comma across all 12 fifths. Each fifth is flattened by $23.46/12 \approx 1.955$ cents. The individual fifths are no longer pure, but the error is small enough to be acceptable — and crucially, it is uniform, so every key sounds equally good (or equally impure, depending on your perspective).

The Pythagorean comma being small — about 1.96% of the octave — is precisely why 12-TET works. It is small because 7/12 is an unusually good convergent of $\log_2(3/2)$. The two facts are the same fact. The comma is the numerator of the error when you approximate $\log_2(3/2)$ by $7/12$, multiplied up by 12 fifths’ worth of accumulation. When the approximation is good, the comma is small, and the distribution is imperceptible. This is why the piano is tuned the way it is.

The group theory

We are now ready for the algebra. In 12-TET, pitch classes form the set $\{0, 1, 2, \ldots, 11\}$ where we identify 0 with C, 1 with C♯, 2 with D, 3 with D♯, 4 with E, 5 with F, 6 with F♯, 7 with G, 8 with G♯, 9 with A, 10 with A♯, and 11 with B. Addition is modulo 12: after 11 comes 0 again, because after B comes C in the next octave (same pitch class). This is $\mathbb{Z}_{12}$, the integers mod 12, and it is a group under addition [1].

Transposition by a semitone is addition of 1. Transposition by a perfect fifth is addition of 7, because the fifth is 7 semitones in 12-TET. Start from C (0) and repeatedly add 7, always reducing modulo 12:

In note names: C, G, D, A, E, B, F♯, C♯, G♯, D♯/E♭, A♯/B♭, F, C. That is the circle of fifths — all 12 pitch classes visited exactly once before returning to the start. The circle of fifths is the orbit of 0 under repeated addition of 7 in $\mathbb{Z}_{12}$.

Why does the orbit visit all 12 elements? Because $\gcd(7, 12) = 1$. This is Bézout’s theorem applied to cyclic groups: an element $g$ generates $\mathbb{Z}_n$ (i.e., its orbit under repeated addition covers all of $\mathbb{Z}_n$) if and only if $\gcd(g, n) = 1$. The generators of $\mathbb{Z}_{12}$ are exactly the elements coprime to 12: that is $\{1, 5, 7, 11\}$. Musically: transposition by 1 semitone (chromatic scale), by 5 semitones (perfect fourth), by 7 semitones (perfect fifth), or by 11 semitones (major seventh) each generates all 12 pitch classes. Transposition by 2 (a whole tone) does not — it produces only the 6-element whole-tone scale. Transposition by 3 (a minor third) produces only the 4-element diminished seventh chord.

This is not a curiosity; it is the algebraic skeleton of tonal music. The circle of fifths closes because 7 and 12 are coprime. That coprimality is guaranteed by the continued fraction structure: the numerator and denominator of a convergent in lowest terms are always coprime (as they must be, being a reduced fraction), and 7/12 is such a convergent.

Now consider the subgroups of $\mathbb{Z}_{12}$. By Lagrange’s theorem, subgroups of a finite group must have orders dividing the group order. The divisors of 12 are 1, 2, 3, 4, 6, and 12, so these are the only possible subgroup orders. For cyclic groups there is exactly one subgroup of each order dividing $n$, and it is generated by $n/d$ where $d$ is the subgroup order. The full list:

The trivial subgroup of order 1 is just $\{0\}$. The subgroup of order 2 is $\{0, 6\}$, generated by 6 — that is, the tritone axis, the interval of exactly half an octave. The subgroup of order 3 is $\{0, 4, 8\}$, generated by 4 — this is the augmented triad, three notes equally spaced around the octave by major thirds. The subgroup of order 4 is $\{0, 3, 6, 9\}$, generated by 3 — the diminished seventh chord, four notes equally spaced by minor thirds. The subgroup of order 6 is $\{0, 2, 4, 6, 8, 10\}$, generated by 2 — the whole-tone scale. And the full group of order 12 is all of $\mathbb{Z}_{12}$.

Each of these has a musical life. The augmented triad ($\{0, 4, 8\}$) sounds ambiguous because it maps onto itself under transposition by a major third — there are only 4 distinct augmented triads total, not 12. Composers exploit this ambiguity when they want harmonic instability without committing to a direction. The diminished seventh ($\{0, 3, 6, 9\}$) is similarly ambiguous: it has only 3 distinct forms and can resolve to any of several keys, which is why it appears so often at structural pivots in Romantic music. These properties are direct consequences of the subgroup structure of $\mathbb{Z}_{12}$.

Messiaen’s modes as cosets

Olivier Messiaen described his “modes of limited transposition” in his 1944 treatise Technique de mon langage musical. He identified seven scales — including the whole-tone scale and the octatonic scale — that have the peculiar property of mapping onto themselves under some transposition strictly smaller than an octave. He found them by ear, by introspection, and by exhaustive search at the keyboard. He did not have the group theory. But the group theory makes their existence not merely explainable but inevitable.

Here is the key definition. A scale $S \subseteq \mathbb{Z}_{12}$ is a mode of limited transposition if there exists some $t \in \{1, 2, \ldots, 11\}$ such that $S + t \equiv S \pmod{12}$ (as a set). In other words, transposing the scale by $t$ semitones maps the scale onto itself. The integer $t$ is called a period of the scale.

Now, the set of all periods of $S$ — together with 0 — forms a subgroup of $\mathbb{Z}_{12}$ (it is closed under addition modulo 12, since if both $t_1$ and $t_2$ are periods then so is $t_1 + t_2$). Call this subgroup $H$. The condition for $S$ to be a mode of limited transposition is simply that $H$ is nontrivial — that is, $H \neq \{0\}$.

Moreover, if $H$ is the period subgroup of $S$, then $S$ must be a union of cosets of $H$ in $\mathbb{Z}_{12}$. This follows immediately from the fact that $H$ acts on $S$ by translation and maps $S$ to itself: every element of $S$ belongs to exactly one coset of $H$, and $S$ is a union of whole cosets. The size of $S$ must therefore be a multiple of $|H|$.

The whole-tone scale $\{0, 2, 4, 6, 8, 10\}$ is itself the unique subgroup of order 6 in $\mathbb{Z}_{12}$. Its period subgroup is the whole-tone scale itself. Transposing by any even number (2, 4, 6, 8, or 10) maps it to itself. Transposing by an odd number gives the complementary whole-tone scale $\{1, 3, 5, 7, 9, 11\}$. There are therefore only 2 distinct transpositions of the whole-tone scale, not 12.

The octatonic (diminished) scale $\{0, 1, 3, 4, 6, 7, 9, 10\}$ has period subgroup $\{0, 3, 6, 9\}$ — the subgroup of order 4. It is a union of two cosets: $\{0, 3, 6, 9\}$ itself and $\{1, 4, 7, 10\}$. Transposing by 3 maps it onto itself. There are only 3 distinct transpositions. Messiaen calls this his Mode 2.

The general formula is clean: a mode of limited transposition with period subgroup of order $d$ has exactly $12/d$ distinct transpositions. For the whole-tone scale, $d = 6$ gives $12/6 = 2$ transpositions. For the octatonic scale, $d = 4$ gives $12/4 = 3$ transpositions.

What Messiaen found by ear was the complete classification of subsets of $\mathbb{Z}_{12}$ that are unions of cosets of a nontrivial subgroup [5]. The group theory makes their existence a theorem rather than a discovery. I find this genuinely beautiful: a composer’s intuition about harmonic symmetry turns out to be an exercise in the theory of cosets of cyclic groups. For the full analysis of each of Messiaen’s seven modes in these terms, see Messiaen, Modes, and the Group Theory of Harmony.

Why not 53?

Given that 53-TET approximates the fifth with an error of less than 0.004% — compared to 12-TET’s 0.11% — one might ask why we do not simply use 53-TET. The mathematical case is overwhelming. In addition to the nearly perfect fifth, 53-TET gives excellent approximations to the just major third (frequency ratio 5:4) and the just minor third (6:5). It was seriously advocated by the 19th-century theorist Robert Holford Macdowall Bosanquet, who even built a 53-key harmonium to demonstrate it. The Chinese theorist Jing Fang described a 53-note system in the 1st century BC. The Arabic music theorist Al-Farabi considered 53-division scales in the 10th century. Everyone who has ever thought carefully about tuning arrives at 53 eventually.

And yet no 53-TET instrument has ever entered widespread use. The reason is anatomical, not mathematical. A piano with 53 keys per octave spans more than 2 metres per octave at any reasonable key size — impossible to play. A guitar with 53 frets per octave has frets spaced roughly 3–4 millimetres apart in the upper register: no human fingertip is narrow enough to press a single fret without touching its neighbours. Even if you could play it, reading 53-TET notation would require an entirely new theoretical and pedagogical apparatus.

The constraint is: we want the largest $N$ such that (a) $N$ is a convergent denominator of $\log_2(3/2)$, so the fifth approximation is genuinely good, and (b) $N$ is small enough to navigate with human hands and readable at a glance. The convergent denominators are 1, 2, 5, 12, 41, 53, 306, … Of these, 12 is the largest that satisfies condition (b). The next convergent, 41, already strains human dexterity — 41-TET keyboard instruments have been built experimentally but never mass-produced. At 53 the case is closed.

One might argue about where exactly the cutoff is, and reasonable people might draw it at 19 or 31 (which are not convergents but have other virtues). But the point is that 12 is not merely a local optimum found by trial and error. It is the specific value where the continued fraction and human physiology intersect.

Closing

There is something I find genuinely satisfying about this argument. Music feels like the most human of activities — expressive, cultural, steeped in history and tradition. And yet the number 12, which lies at the foundation of so much of the world’s music, is not a human choice at all. It is the continued-fraction convergent of an irrational number that was fixed by the physics of vibrating strings long before any human struck a tuning fork.

The circle of fifths closes because $\gcd(7, 12) = 1$: a fact about integers, not about culture. Messiaen’s modes exist because $\mathbb{Z}_{12}$ has nontrivial proper subgroups: a fact about cyclic groups, not about 20th-century French aesthetics. The augmented triad sounds ambiguous because it is a coset of the order-3 subgroup of $\mathbb{Z}_{12}$: a fact about quotient groups, not about Romantic harmony conventions.

I came to music theory sideways — through acoustics, then signal processing, then the mathematics of scales. What surprised me, when I finally worked through the continued fraction argument properly, was not that the math existed but that it was so tight. There is essentially no freedom in the answer. Given the constraint that a musical scale should be built around the most consonant interval (after the octave), should form a closed group structure, and should be navigable by a human performer, the answer is 12. Not approximately 12, not 12 as a historical compromise. Exactly 12.

The number is not a tradition. It is a theorem.

For more on related themes: the Fibonacci sequence and golden ratio in music appear in Fibonacci, Lateralus, and the Golden Ratio. The Euclidean algorithm and rhythmic structure are explored in Euclidean Rhythms — a sister post to this one in the math-and-music thread. And for the physics of audio sampling rates, where a similar interplay of number theory and practical constraints forces another specific number, see Why 44,100 Hz?.

References

[1] Balzano, G. J. (1980). The group-theoretic description of 12-fold and microtonal pitch systems. Computer Music Journal, 4(4), 66–84.

[2] Carey, N., & Clampitt, D. (1989). Aspects of well-formed scales. Music Theory Spectrum, 11(2), 187–206.

[3] Milne, A., Sethares, W. A., & Plamondon, J. (2007). Isomorphic controllers and dynamic tuning. Computer Music Journal, 31(4), 15–32.

[4] Lloyd, L. S., & Boyle, H. (1978). Intervals, Scales and Temperaments. St. Martin’s Press.

[5] Douthett, J., & Steinbach, P. (1998). Parsimonious graphs: A study in parsimony, contextual transformations, and modes of limited transposition. Journal of Music Theory, 42(2), 241–263.

Changelog

• 2025-11-20: Updated the spelling of “Robert Holford Macdowall Bosanquet” (previously rendered as “Macdowell”).
• 2025-11-20: Changed “about 1.36% of the octave” to “about 1.96% of the octave.” The 1.36% figure is the frequency ratio above unity (531441/524288 ≈ 1.01364); the logarithmic fraction of the 1200-cent octave is 23.46/1200 ≈ 1.96%.
• 2025-11-20: Changed “12 octaves’ worth of accumulation” to “12 fifths’ worth of accumulation.” The Pythagorean comma accumulates over 12 stacked fifths (which span approximately 7 octaves), not 12 octaves.

Physics training means coming to mathematics as a tool before arriving at it as an object of aesthetic interest, and it took me longer than it should have to notice that a proof can be beautiful in the same way a piece of music can be beautiful — not despite its rigour but because of it. Both reward attention to structure. Both have surfaces accessible to a casual listener and depths that only reveal themselves when you look harder.

Lateralus, the title track of Tool’s 2001 album, is a convenient case study for the overlap. It is not the only piece of music built around Fibonacci numbers — Bartók made the connection decades earlier, and it appears in scattered places across Western and non-Western traditions — but it is among the most thoroughly and deliberately constructed, and the mathematical structure is audible rather than merely theoretical.

What follows is an attempt to do justice to both dimensions: the mathematics of the Fibonacci sequence and the golden ratio, and the musical mechanics of how those structures show up and what they do.

The Fibonacci Sequence

The sequence is defined by a recurrence relation. Starting from the initial values $F(1) = 1$ and $F(2) = 1$, each subsequent term is the sum of the two preceding ones:

This gives:

The term $987$ is the sixteenth Fibonacci number, $F(16)$. Keep that in mind.

The recurrence can be encoded compactly in a matrix formulation. For $n \geq 1$:

This is more than notational tidiness — it connects the Fibonacci sequence to the eigenvalues of the matrix $\mathbf{A} = \bigl(\begin{smallmatrix}1 & 1 \\ 1 & 0\end{smallmatrix}\bigr)$, which are exactly $\varphi$ and $-1/\varphi$ where $\varphi$ is the golden ratio. That connection gives us Binet’s formula, a closed-form expression for the $n$-th Fibonacci number:

Since $|\psi| < 1$, the term $\psi^n / \sqrt{5}$ diminishes rapidly, and for large $n$ we have the convenient approximation:

This means Fibonacci numbers grow exponentially, at a rate governed by the golden ratio. The sequence does not grow linearly or polynomially; it spirals outward.

The Golden Ratio

The golden ratio $\varphi$ appears as the limit of consecutive Fibonacci ratios:

It can be derived from a simple geometric proportion: divide a line segment into two parts such that the ratio of the whole segment to the longer part equals the ratio of the longer part to the shorter part. Calling those ratios $r$:

What makes $\varphi$ mathematically distinctive is its continued fraction representation:

This is the simplest possible infinite continued fraction. It is also, in a precise sense, the hardest real number to approximate by rational fractions. The convergents of a continued fraction are the best rational approximations to a real number at each level of precision; the convergents of $\varphi$ are exactly the ratios of consecutive Fibonacci numbers: $1/1$, $2/1$, $3/2$, $5/3$, $8/5$, $13/8$, $\ldots$ These converge more slowly to $\varphi$ than the convergents of any other irrational number. $\varphi$ is, in this sense, maximally irrational.

That property has a physical consequence. In botanical phyllotaxis — the arrangement of leaves, seeds, and petals on plants — structures that grow by adding new elements at a fixed angular increment will pack most efficiently when that increment is as far as possible from any rational fraction of a full rotation. The optimal angle is:

This is the golden angle, and it is the reason sunflower seed spirals count $55$ and $89$ (consecutive Fibonacci numbers) in their two counter-rotating sets. The mathematics of efficient growth in nature and the mathematics of the Fibonacci sequence are the same mathematics.

The golden spiral — the logarithmic spiral whose growth factor per quarter turn is $\varphi$ — is the visual representation of this: it is self-similar, expanding without bound while maintaining constant proportionality.

Fibonacci Numbers in Music: Before Tool

The connection between the Fibonacci sequence and musical structure is not Tool’s invention. The most carefully documented case is Béla Bartók, whose Music for Strings, Percussion and Celesta (1936) has been analysed exhaustively by Ernő Lendvai. In the first movement, the climax arrives at bar 55 (a Fibonacci number), and Lendvai counted the overall structure as 89 bars — the score has 88, but he added an implied final rest bar to reach the Fibonacci number — dividing at bar 55 with near-mathematical precision. Lendvai argued that Bartók consciously embedded Fibonacci proportions into formal structure, tonal architecture, and thematic development throughout much of his output.

Whether these proportions were conscious design or an instinct that selected naturally resonant proportions is contested. The same question applies to claims about Mozart and Chopin. What is more defensible is a structural observation about the piano keyboard and Western scales that requires no attribution of intent:

A single octave on the piano keyboard has 13 keys, comprising 8 white keys and 5 black keys. The black keys are grouped as 2 and 3. The numbers $2, 3, 5, 8, 13$ are five consecutive Fibonacci numbers — $F(3)$ through $F(7)$.

The standard Western scales make this concrete. The major scale contains 7 distinct pitches within an octave of 12 semitones. The pentatonic scale (ubiquitous in folk, blues, rock) contains 5 pitches. The chromatic scale contains 12 pitch classes per octave; counting both endpoints of the octave (C to C) gives 13 chromatic notes, the next Fibonacci number.

Harmonic intervals in just intonation are rational approximations of simple frequency ratios: the octave (2:1), the perfect fifth (3:2), the perfect fourth (4:3), the major third (5:4), the minor third (6:5). The numerators and denominators are small integers, often Fibonacci numbers or their neighbours. The major triad — the structural foundation of tonal Western music — consists of intervals in frequency ratios $4:5:6$, three consecutive integers that bracket the Fibonacci-adjacent range.

This does not mean that Western music is secretly Fibonacci. It means that the integer frequency ratios that produce consonant intervals are the small integers, and small integers include the small Fibonacci numbers. The connection is genuine but not exclusive.

Lateralus

Tool’s Lateralus (2001, album of the same name) is unusual in that the Fibonacci construction is not an analytical inference applied after the fact — it was discussed publicly by the band. Drummer Danny Carey has spoken about his engagement with sacred geometry and mathematical structure, and the song’s construction has been described as intentional by multiple band members.

There are two primary levels of Fibonacci structure in the song. The third — the thematic content of the lyrics — makes the mathematical frame explicit.

The Syllable Count

The opening verses are constructed so that successive lines contain syllable counts following the Fibonacci sequence ascending: $1, 1, 2, 3, 5, 8, 13$. The first syllable count is a single word. The second is another. The third is a two-syllable phrase. The sequence continues, each line adding the weight of the previous two, until the thirteenth-syllable line, which in structure and delivery feels like the crest of a wave.

The second half of the verse then descends: $13, 8, 5, 3, 2, 1, 1$. Or, in some analyses, the chorus and pre-chorus sections begin a new ascending Fibonacci run before the full descent, creating a nested structure of expansions and contractions.

The audible effect of this design is not arbitrary. A sequence of lines whose syllable counts follow $1, 1, 2, 3, 5, 8, 13$ creates a consistently accelerating density of text over the same musical time. The vocal line becomes more compressed as the syllable count rises, building tension — and then the descent releases it. This is not how most pop or rock lyrics are structured. It produces a breathing, organic quality, the way a plant reaches toward light.

The Time Signature: 987

The verse sections of the song cycle through three time signatures in succession: $9/8$, then $8/8$, then $7/8$.

This three-bar pattern repeats. Now: the sequence of numerators is $9$, $8$, $7$. Written as a three-digit number: 987. And as noted above, $987 = F(16)$, the sixteenth Fibonacci number.

Whether this is a deliberate encoding or a remarkable coincidence is a matter of interpretation. The time signature sequence is definitely deliberate — asymmetric meters of this kind require careful compositional choice. The fact that their numerators concatenate to a Fibonacci number is either intentional and clever or accidental and still remarkable. Either way, the time signature pattern has a musical function independent of the Fibonacci reading.

In standard rock, time is almost always $4/4$: four even beats per bar, a pulse that is maximally predictable and maximally amenable to groove. The $9/8 + 8/8 + 7/8$ pattern is the opposite. Each bar has a different length. The listener’s internal metronome, calibrated to $4/4$, cannot lock onto the pattern. The music generates forward momentum not through a repeated downbeat but through the continuous, non-periodic unfolding of measures whose lengths shift. This is the rhythmic analogue of a spiral: no two revolutions are identical in length, but the growth is consistent.

The chorus and other sections use different time signatures, including stretches in $5/8$ and $7/8$ — Fibonacci numbers again, and specifically the $5, 8, 13$ triplet that appears so often in this context.

The Thematic Content

The lyrics are explicitly about spirals, Fibonacci growth, and the experience of reaching beyond a current state of development. They reference the idea of expanding one’s perception outward through accumulating cycles, each containing and exceeding the previous one. The chorus refrain — about spiralling outward — names the mathematical structure of the golden spiral directly. The song is, in its own terms, about the process that the mathematics describes.

This kind of thematic coherence between structure and content is what makes the construction interesting rather than merely clever. The Fibonacci structure is not decorative. It is the argument of the song made manifest in its form.

Why Fibonacci Structure Works in Music

The most interesting question is not whether the Fibonacci structure is there — it clearly is — but why it produces the musical effect it does.

Consider what the Fibonacci sequence represents physically. It is the growth law of structures that build on their own preceding state: $F(n) = F(n-1) + F(n-2)$. Unlike arithmetic growth (add a constant) or geometric growth (multiply by a constant), Fibonacci growth is self-referential. Each term contains the memory of the previous two. The sequence is expansive but not uniform; it accelerates, but always in proportion to what came before.

Musical tension and release are, in an important sense, the same mechanism. A phrase creates an expectation; its continuation either confirms or subverts that expectation; resolution reduces the tension. What makes a musical phrase feel like it is building toward something is precisely the progressive accumulation of expectation — each bar adding its weight to the previous, the accumulated tension requiring resolution at a scale proportional to the build-up. The Fibonacci syllable structure in Lateralus generates this literally: each line is denser than the previous two lines’ combined syllable count would suggest is comfortable, until the structure has to breathe.

The time signature asymmetry works similarly. In $4/4$, the beat is predictable, and the listener’s body can lock to it and then coast on that lock. In $9/8 + 8/8 + 7/8$, the beat is never fully locked — the pattern is periodic (it repeats) but the internal structure of each repetition is shifting. The listener is perpetually catching up, perpetually leaning slightly into the music to find the next downbeat. This is not discomfort — it is engagement. The mathematical reason is that the pattern is large enough to be periodic (it does repeat) but small enough to be audible as a unit. The brain can learn the 24-beat super-pattern; it just requires attention that $4/4$ does not.

There is a deeper reason why golden-ratio proportions feel right in musical form. The golden section of a piece — the point at which the piece divides in the $\varphi : 1$ ratio — is the point of maximum accumulated development before the final resolution. In a five-minute piece, the golden section falls at roughly 3:05. This is, empirically, where the emotional and structural climax tends to sit in a wide range of well-regarded music, from Baroque to jazz. Whether composers consciously target this proportion or whether the proportion is what accumulated development looks like when done well is not easily separable. But the mathematical reason it is a proportion worth targeting is that $\varphi$ is the only division point that is self-similar: the ratio of the whole to the longer part equals the ratio of the longer part to the shorter part. There is no arbitrary scale associated with the golden section; it is scale-invariant, the same proportion at every level of analysis.

A Brief Note on Binet and Limits

The closed-form expression for Fibonacci numbers,

has a pleasing consequence for large $n$. Since $|\psi| \approx 0.618 < 1$, the term $\psi^n \to 0$, and $F(n)$ is simply the nearest integer to $\varphi^n / \sqrt{5}$. The integers produced by the Fibonacci recurrence are the integers that $\varphi^n / \sqrt{5}$ passes closest to. The exponential growth of $\varphi^n$ and the rounding to integers together give the sequence.

This is also why the ratios $F(n+1)/F(n)$ converge to $\varphi$ exponentially fast — the error is $\mathcal{O}(|\psi/\varphi|^n) = \mathcal{O}(\varphi^{-2n})$ — and why, for musical purposes, the Fibonacci ratios $8:5$, $13:8$, $21:13$ are already excellent approximations of the golden ratio, close enough that the ear cannot distinguish them from $\varphi$ in any direct sense.

What Lateralus Is

Lateralus is not a math lecture set to music. It is a nine-minute progressive metal track that is physically involving, rhythmically complex, and lyrically coherent. The Fibonacci structure would be worthless if the song were not also, on purely musical terms, good.

What the mathematics adds is a vocabulary for something the song achieves anyway: the sense of growing without ever arriving, of each section being both a resolution of what came before and an opening toward something larger. The golden spiral does not end. The Fibonacci sequence does not converge. The song does not resolve in the sense that a classical sonata resolves; it spirals to a close.

The reason this is worth writing about is that it makes concrete a connection that is usually stated vaguely: mathematics and music are similar. They are similar in specific and articulable ways. The self-referential structure of the Fibonacci recurrence, the scale- invariance of the golden ratio, the information-theoretic account of tension and expectation — these are not metaphors for musical experience. They are, in this case, the actual mechanism.

References

Lendvai, E. (1971). Béla Bartók: An Analysis of His Music. Kahn & Averill.

Benson, D. J. (2006). Music: A Mathematical Offering. Cambridge University Press. (For an introduction to the general theory of tuning, temperament, and harmonic series.)

Tool. (2001). Lateralus. Volcano Records.

Livio, M. (2002). The Golden Ratio: The Story of Phi, the World’s Most Astonishing Number. Broadway Books.

Knott, R. (2013). Fibonacci numbers and the golden section in art, architecture and music. University of Surrey Mathematics Department. https://r-knott.surrey.ac.uk/Fibonacci/fibInArt.html

Changelog

• 2025-11-20: Clarified the Bartók bar count: the written score has 88 bars; Lendvai’s analysis counted 89 by adding an implied final rest bar to reach the Fibonacci number. Previously stated as “89 bars” without qualification.