The music of the primes · part 1 of 3

Primes tame quantum chaos

An expedition across 17 orders of magnitude in height — from the first zeros of the Riemann zeta function to zeros whose index numbers have 22 digits. Everywhere we find the fingerprints of the primes, which no randomness can imitate.

David Hessabout 9 min readnumerical number theory
Re(s) = ½ · critical lineIm(s) →14.1321.0225.0130.4232.9437.5940.9243.3348.0149.77

Imagine the chaos inside an atomic nucleus and the perfect order of the prime numbers dancing on the same floor. It sounds like a sci-fi script, yet this is exactly what our latest numerical study shows.

We examined the zeros of the Riemann zeta function over an enormous range: from the first few thousand up to heights around 1021, where zeros carry index numbers with 22 digits. The question was simple — do the zeros behave like random chaos, or do they carry a memory of the primes? The answer is clear: the primes leave fingerprints in the distribution of the zeros that no randomness can reproduce.

17orders of magnitude in height on the critical line
18independent data blocks
≤ 1.21%deviation of Berry's formula from the data
0free parameters to tune
The range of the expedition
From height 10⁴ to 10²¹ — like comparing the length of a table with the distance to the nearest stars.
104108101210161021Start: t ≈ 10⁴the first few thousand zerosFinish: t ≈ 10²¹zero index numbers have 22 digitsheight t on the critical line (log axis) · 17 orders · 18 data blocks
Quick primer · what is a zeta zero

The zeta function ζ(s) = 1 + 1/2s + 1/3s + … is tied to the primes so tightly that in 1859 Bernhard Riemann showed how its zeros can exactly “play back” the distribution of primes. The non-trivial zeros have the form ½ + i·γ — they lie on the so-called critical line (the Riemann Hypothesis says all of them do). The number γ is their “height” t. You can see the first ten in the header of this article.

01Quantum chaos and atomic nuclei: why do physicists care about pure maths?

In the early 1970s the mathematician Hugh Montgomery computed how often two zeta zeros occur at a given distance from each other. When he showed the formula to the physicist Freeman Dyson, Dyson immediately recognised a familiar face: this is exactly the eigenvalue statistics of large random matrices — specifically of the GUE (Gaussian Unitary Ensemble).

Random-matrix theory made its name describing the energy levels of heavy atomic nuclei, where quantum chaos rules. So at short distances the zeta zeros behave like the levels of a chaotic quantum system: they repel each other and almost never sit right next to one another.

Sounds great. But there is a catch.

Random matrices describe local chaos — how zeros repel at close range. As soon as we look at long distances and count how many zeros fit into a wider window, random-matrix theory fails. It predicts that the fluctuation in the number of zeros keeps growing without limit. In reality the growth stops and saturates.

Picture this · cars on a motorway

You stand on a bridge over a motorway and count the cars passing through a stretch of length L. Drivers keep their distance (that is the “repulsion”). Each time you get a slightly different count — and the variance of that count is what we call Σ²(L). For random matrices it grows forever with the length of the stretch. For zeta zeros it levels off at some point, as if the traffic were run by a hidden conductor. That conductor is the primes.

Counting zeros in a window: chaos vs. primes
Number variance Σ²(L) in a window of length L. Top: height t ≈ 10⁴ (H ≈ 7.4); bottom: t ≈ 10²¹ (H ≈ 46.5). Hover over the chart.
Berry: primes as periodic orbitspure GUE (random matrices)
0.00.20.40.60.8010203040window length L (in mean spacings)Σ²(L)t ≈ 10⁴ · H ≈ 7.4saturation ≈ 0.34peaks from the prime 2spacing H/ln 2 ≈ 10.6GUE keeps growing like ln L …
0.00.20.40.60.81.004080120160window length L (in mean spacings)Σ²(L)t ≈ 10²¹ · H ≈ 46.5higher plateau than above: more primes fit below the Heisenberg timeGUE keeps growing …
Theoretical curves computed to illustrate the shape: Berry's diagonal formula (all prime powers with period ln pᵏ shorter than the Heisenberg time H = ln(t/2π)) against pure GUE random-matrix theory. These are not the study's measured data — those agree with Berry's formula to 0.26–1.21%.

02Berry's magic: primes as orbits

In 1988 the British physicist Sir Michael Berry came up with an elegant idea. We can think of the zeta zeros as the energy levels of a (still unknown) quantum system — and of the primes as its closed periodic orbits. Each prime p and its powers contribute an “orbit” with period ln pk.

From this picture Berry derived a semiclassical formula for the number variance Σ²(L) that has no free parameters. Nothing is tuned, nothing is fitted to the data. The formula predicts that the variance first oscillates and then saturates.

The study tested it systematically on 18 data blocks across 17 orders of height:

  • Accuracy: Berry's formula matches the real data with a deviation of just 0.26–1.21%.
  • The signature of the prime 2: the variance chart shows fine ripples. They come from the shortest prime orbit — the prime 2 — with period ≈ H/ln 2. At height t ≈ 104 that is about 10.6 mean spacings (see the top chart above).

03The big test: real primes vs. a random fake

You might object: “Isn't it all just because there are a certain number of primes and they get sparser?” To rule this out we pitted Berry's formula against the Cramér model of random primes — fictitious “primes” with the same density as the real ones but random positions.

The big test: real primes vs. a fake
How far is each theory from the real zeros? Further left = better agreement.
0.1 %1 %10 %100 %deviation from the measured variance Σ²(L) (log axis)Real primesBerry's formula0.26–1.21 %Random “primes”Cramér model5.4–9.9 %No primespure GUE15–82 %
Log axis: each tick is ten times the previous one. Real and random primes differ by roughly an order of magnitude.

The lesson: the exact positions of the primes matter. Even at height 1021 the zeta zeros “remember” the specific arithmetic sequence 2, 3, 5, 7, 11, … — the right count alone is not enough.

04Saturation is not constant: height changes the rules

It was long quietly assumed that after suitable rescaling (so-called unfolding, which sets the mean spacing between zeros to 1) the zeros behave the same at every height. For long-range variance this is not true.

0.34saturation level for the lowest zeros (t ≈ 10⁴, H ≈ 7.4) — only about 250 primes lie below the Heisenberg time
→ 0.56saturation level at height t ≈ 10²¹ (H ≈ 46.5) — roughly 3.5 quintillion primes lie below the Heisenberg time

Why? The higher you climb, the more primes fit below the so-called Heisenberg time H = ln(t/2π) — that is, the more orbits have a period ln p shorter than H. Each extra prime adds a little to the height of the plateau. But it grows incredibly slowly, roughly like the logarithm of the logarithm of the height: 17 orders of magnitude add only about two tenths.

05A four-stage sieve: how to recognise the true zeta spectrum?

To stress-test the findings we built a four-stage sieve — a kind of lie detector for candidate spectra:

  • 1 · Density and rigidity: do the total level count and its stability match?
  • 2 · GUE spacings: do neighbouring levels behave locally like random matrices?
  • 3 · Prime content: does the spectrum contain prime oscillations?
  • 4 · Saturation: does the growth of the variance stop at long range?
Spectral model1 · Density & rigidity2 · GUE spacings3 · Prime oscillations4 · Saturation
Pure GUE random matrix
no arithmetic
✓ passes✓ passes✗ fails✗ fails
Actual zeta zeros✓ passes✓ passes✓ passes✓ passes
Levels of the xp model + prime orbits✓ passes✓ passes✓ passes✓ passes

A pure random matrix passes the first two stages but fails on primes and saturation — it contains no arithmetic. Both the actual zeros and the theoretical levels of the Berry–Keating xp model supplemented with prime orbits pass all four. That is a valuable guide to what the sought-after quantum “operator” would have to look like — if, as the Hilbert–Pólya conjecture suggests, it exists at all.

06What follows (and what does not)

✓ What the study showed

Berry's semiclassical theory describes the behaviour of the zeros across 17 orders of height without a single free parameter — and the real primes in it cannot be replaced by random ones. All data and scripts are published under an open licence with SHA-256 checksums, so anyone can reproduce the computations.

✗ What does not follow

The work makes no claim about the Riemann Hypothesis. It does show, however, that anyone wanting to understand the zeta function has to take chaos physics and the primes together — one without the other is not enough.

Glossary
GUE
Gaussian Unitary Ensemble — random Hermitian matrices. Their eigenvalues repel each other exactly like zeta zeros at short distances.
Unfolding
Rescaling the zeros so that the mean spacing between neighbours is 1. Only then can zeros from different heights be compared.
Σ²(L) — number variance
How much the number of zeros in a window of length L fluctuates around its mean L.
Heisenberg time H
Here H = ln(t/2π). Orbits (primes) with period ln p shorter than H govern the long-range behaviour of the zeros.
Cramér model
Random “primes”: each number n is chosen with probability 1/ln n. The density is right, the positions are random.
Hilbert–Pólya conjecture
The idea that the zeta zeros are the eigenvalues of some self-adjoint operator — a quantum system nobody has found yet.
Paper, data and code

Study I · Berry saturation of the Riemann zeros

The whole series on Zenodo: Study I · Study II · Study III · licence CC BY 4.0 (text, data) and MIT (code)

The charts are interactive — hover over the curves and points.

2 komentáře

Napsat komentář

Vaše e-mailová adresa nebude zveřejněna. Vyžadované informace jsou označeny *