The Stress Test: Stability, Decay, and the Trivial Zeros

The two trials

A mental model earns its keep by surviving pressure, so here is the pressure. Physical reality imposes two demands that any candidate account of matter has to meet: stability, meaning a spectrum of allowed, persistent states, and decay, meaning that states leak, radiate, and end. The model of the previous chapters says that a non-trivial zero is the terminating value where the field localises. If that reading has content, then the arrangement of the zeros should say something about how stable states organise, and leaving the critical line should correspond to losing stability.

Both trials are run below with the register attached. The first goes better than the model deserves. The second is more interesting, because it is where the analogy is doing the most work and where the least of it is secured.

Trial one: spacing, spectral repulsion, and heavy nuclei

If a zero marks a localised state, the intervals between zeros should behave like the intervals between levels of a real spectrum. They do, and this is the strongest piece of external support the whole line of thought has. Montgomery's 1972 pair-correlation calculation for the zeros produced a form that Dyson recognised on sight as the pair correlation of eigenvalues of a large random Hermitian matrix from the Gaussian Unitary Ensemble — the ensemble that describes the level statistics of heavy nuclei such as erbium-168 and uranium-238, and of chaotic quantum systems with broken time-reversal symmetry. Odlyzko's numerical work, carried out at heights where millions of zeros can be compared against the prediction, confirmed the agreement to a precision that leaves little room for coincidence.

The mechanism on the physical side has a name: level repulsion. Nuclear levels do not cluster at random because the states interact; the eigenvalues of a Hermitian operator push apart, and the resulting distribution has a hole at zero spacing that a Poisson process does not. The zeros of ζ show exactly this repulsion. That is not a poetic resemblance; it is the same distribution, tested numerically, and it is the reason the Hilbert–Pólya conjecture — that the zeros are the eigenvalues of some self-adjoint operator — remains the most respectable guess about why the critical line should be a line at all.

The recursive side supplies the anchor. In the Berry–Keating semiclassical reading, the primes play the role of the periodic orbits of the putative classical system, entering the explicit formula as log-spaced frequencies while the zeros appear as the spectrum. Read through the volume's vocabulary: the recursive input, the multiplicative structure of the primes, fixes the statistical layout of the iterative output, the sequence of zeros — and that layout is the one physical nuclei display.

Verdict on trial one: the model serves, and it serves because someone else did the work. What is established is a shared distribution and an unproven but well-motivated conjecture about a shared operator. What is not established is that the states being spaced are masses. The spacing statistics constrain the arithmetic of ζ and the arithmetic of matrices; they say nothing yet about kilograms.

Trial two: drifting off the critical line

Here is the sharper test. Suppose a zero were found with real part σ ≠ 1/2. In mathematics that retires the Riemann hypothesis. In the model it should mean something specific, and it does have a natural reading: the loss of a balance that was holding a state still.

The wave-mechanical grammar behind that reading is exact and worth writing out, because it is the only part of the analogy that is more than a gesture. A mode with a complex frequency carries a factor e^{iωt} with ω = ωᵣ + iγ; the imaginary part becomes a real exponential envelope e^{−γt}, and that is precisely how physics writes a decaying resonance — a Breit–Wigner line with a width proportional to the inverse lifetime, the standard description of an unstable nucleus or a short-lived particle. A state is stable when the envelope's exponent vanishes and unstable when it does not. Displacement off a line in a complex plane is therefore the right shape for the physics of decay, and the model is not inventing that correspondence.

The functional equation then supplies a genuine symmetry to reason with. ξ(s) = ξ(1 − s) forces zeros to occur in the configuration s, 1 − s, and with complex conjugation, in quadruples off the line. Any drift to the left of 1/2 is accompanied by an exact mirror drift to the right. That is a real theorem about ζ, and it means the model cannot have one-sided decay: whatever leaks on one side is matched on the other. As a bookkeeping picture it is elegant, and the temptation to read the pair as a matter–antimatter ledger, with the balance enforcing conservation, is obvious.

The temptation must be refused, and precisely. The pairing in ξ(s) = ξ(1 − s) is a reflection in the real part; particle–antiparticle conjugation is CPT, an operation on charge, parity, and time reversal in a quantum field theory with a Hilbert space and an S-matrix. No map has been exhibited between them, and the words balanced and symmetric are doing all the connecting. Worse, the observed matter–antimatter asymmetry of the universe is a measured imbalance — baryon-to-photon ratio of order 10⁻⁹ — so a model whose central symmetry is exact would have to explain how the asymmetry arises, and this one has nothing to say about it. The reading is a figure. Filed as such.

There is a second and more mundane objection. ζ has no zeros at all in the region Re(s) > 1, and the classical zero-free region pushes the boundary further; the critical strip is 0 < Re(s) < 1 and the interesting behaviour is confined there. So the diagram of a state drifting outward until it radiates away has no arbitrarily large elsewhere to drift into. Whatever instability means in this vocabulary, it lives inside a strip of width one.

Where the model breaks: zeros as static objects

The stress test locates one clean break. If the zeros are treated as the particles themselves — static points on a line, one per object — the model fails immediately, and it fails for reasons already recorded in Chapters 27 through 29: no operator maps a complex coordinate to a mass, a density matrix, or a field configuration; the thermodynamic reading of ζ does not survive continuation past Re(s) = 1; and the count of zeros below a height is fixed by the Riemann–von Mangoldt formula, which is arithmetic and knows nothing about a particle census.

The version that survives the pressure is weaker and better. The zeros are not the objects; they are the allowed frequencies of the medium — the spectrum, in the same sense that the modes of a cavity are not the light in it. Matter, in a fluid reading, is never a static point but a persistent circulation: a vortex, a soliton, a standing pattern that holds itself together while the medium keeps moving. The zeros, on this reading, say which patterns are permitted to persist; they do not enumerate what is present. That is exactly the shift the spacing statistics of trial one already suggested, since level statistics are a property of a spectrum and not of an inventory.

So the model does not survive as an ontology and does survive as a spectral placeholder. That is the honest outcome of a stress test, and it is a better position than the one it started from.

A ruling on the trivial zeros

The question was whether the trivial zeros at s = −2, −4, −6, … are unmanifested dimensions of the fluid or an artifact to discard. They are neither, and the correct answer is more useful than either option.

They are bookkeeping produced by the completion of the function. The functional equation is cleanest in the form π^{−s/2} Γ(s/2) ζ(s) = π^{−(1−s)/2} Γ((1−s)/2) ζ(1−s), and the gamma factor Γ(s/2) has poles at s = 0, −2, −4, …. ζ must vanish at the negative even integers to keep the product finite. The trivial zeros exist because of the archimedean factor — the gamma function — that has to be attached to ζ before the reflection symmetry can be stated at all. They carry information: they are the signature of the real place in the completed Ξ function, and in the standard treatment they are precisely the zeros that are not on the critical line and are not supposed to be, which is why the Riemann hypothesis is stated about the non-trivial ones.

Translated into the model's own terms: the trivial zeros are not hidden dimensions and not garbage. They are the constraints imposed by the completion — the terms that make the symmetry s ↔ 1 − s legal in the first place. If the model wants a physical name for them, the only defensible one is a normalisation or a boundary condition: the price of stating the reflection, not a set of states hiding behind it. Calling them dark unmanifested dimensions would install new physical content on the strength of a word, which is the failure mode this volume exists to avoid. Discarding them would be worse, because without the gamma factor the symmetry the whole model leans on cannot be written down.

The ruling, and what would change it

The model passes trial one on borrowed strength: the GUE agreement between the zeros and nuclear level statistics is real, tested numerically, and the best reason to keep speaking of the zeros as a spectrum. It passes trial two only in grammar: complex frequencies do produce exponential envelopes, and the functional equation does force mirrored pairs, but no map from a real part to a lifetime and no map from the reflection to CPT has been exhibited, so the decay reading remains an analogy inside wave mechanics rather than a statement about nuclei.

It breaks as an ontology of static points, and the break is the chapter's most useful product: the zeros are better held as the resonant frequencies of the medium that permit stable circulations, not as the circulations themselves.

What would move any of this from figure to result: a self-adjoint operator whose spectrum is the non-trivial zeros, which would convert the Hilbert–Pólya conjecture into a theorem and put the whole vocabulary on a different footing; or, on the physical side, an exhibited map with units — real part to inverse lifetime, imaginary part to energy — tested against a measured resonance width. Absent either, the correct standing is the one this chapter ends with: a stress-tested placeholder that survived one trial, half-passed the second, and shed the reading that could not hold.