# Phase-like behaviour that is not movement of zeros

10 November 2026, America/Los_Angeles. Grant asked to investigate *e* with respect to phase-like behaviour that is not movement of zeros. The interpretation taken here is *e* as in \(e^{i\theta}\): phase.

(Pruned 11 Nov 2026: cross-domain comparison passages removed at Grant's request; original in /workspace/math/pruned-20261111/notes-orig/.)

**Labels.** [known] = established mathematics, with a source. [project] = a proof reconstruction, a certified finite computation, or a numerical check here; not a new number-theory theorem. [speculative] = an unproved additional idea.

This note does not claim RH progress. It was not deployed.

## 0. Verdict

In four precise senses, a “phase” is not the same thing as a movement of zeros.

1. **[known]** For a local factor \(1-a_p p^{1-2s}\) with \(a_p=r e^{i\theta}\), the angle \(\theta\) shifts the column vertically by \(\theta/(2\log p)\). A full \(2\pi\) turn raises every labelled zero by one slot \(\pi/\log p\) and leaves the column, as a set, unchanged: monodromy, or relabelling, not a rearrangement. The radius \(r>0\) is the sideways rigid motion, and has no monodromy.
2. **[known]** Multiplying coefficients by \(n^{-i\tau}\) is the substitution \(s\mapsto s+i\tau\), which shifts *every* zero of a Dirichlet series vertically by \(-\tau\). That is an exact, trivial, rigid phase motion. A Dirichlet character \(\chi(p)=e^{2\pi i k/q}\) is also a phase at the level of coefficients, but it is periodic in \(n\), not a power of \(n\), so \(L(s,\chi)\) is a genuinely different function. Its zeros are not a translate of those of \(\zeta\).
3. **[project]** The completed order zeta for \(D=30\) has a real Hardy function on the half-line, whose phase comes from the Voight completion. Inherited zeros on the quarter-line follow ordinary Gram’s law at half height. Through ordinary height 200 that law holds in every interval; through 2,000 it holds in 1,358 of 1,516 intervals, with first failure the classic empty-then-double block at \(t=280.80\). Local zeros on the half-line do *not* obey a useful Gram law for this completion: the completing phase winds like \(2t\log t\), the zeros only like \((\log 30)\,t\). Through height 1,000, 3,114 of 4,115 Gram intervals are empty.
4. **[known / project]** In the quaternions, \(e^{I\theta}\) rotates the imaginary unit. A zero sphere is the orbit of a complex zero under that action: a phase *symmetry*, not a path. The same lesson as the Q8 Euler product: length is arithmetic, direction is bookkeeping.

## 1. The complex dial: winding versus sliding

**[known]** The zeros of \(1-a_p p^{1-2s}\), for \(a_p=r e^{i\theta}\) with \(r>0\) and \(m\in\mathbb{Z}\), are

\[
s_m(r,\theta)=\frac12+\frac{\log r}{2\log p}+\frac{i(\theta+2\pi m)}{2\log p}.
\]

The real part depends only on \(r\); the height depends on \(\theta\) and the slot index \(m\). In particular the vertical shift is exactly \(\theta/(2\log p)\), and the spacing remains \(\pi/\log p\).

A turn \(\theta\mapsto\theta+2\pi\) sends \(s_m\) to \(s_{m+1}(\,\cdot\,,\theta)\). Every labelled zero climbs one slot. The set of heights is \(\frac{\theta}{2\log p}+\frac{\pi}{\log p}\mathbb{Z}\), which is unchanged. This was checked for \(p=2,3,5\): the \(m=1\) zero at \(\theta=2\pi\) sits exactly where the \(m=2\) zero sat at \(\theta=0\).

The radius is different. Changing \(r\) moves the whole column to a different vertical line, with \(r=p^{-1/2}\) landing on the quarter-line and \(r=p^{1/2}\) on the three-quarter line, as in `zero-movement.md`. The positive ray \(r>0\) is simply connected, so carrying \(r\) around and back to 1 has no monodromy.

**What “phase-like but not movement” means here.** Under \(\theta\), the set is invariant of period \(2\pi\); only the labels cycle. That is relabelling, not dynamics of the set. Under \(r\), the set does move, rigidly and sideways.

## 2. Two kinds of coefficient phase

**[known]** If \(f(s)=\sum a_n n^{-s}\), then \(\sum a_n n^{-i\tau}n^{-s}=f(s+i\tau)\). Every zero \(\rho\) of \(f\) becomes \(\rho-i\tau\). The first eight ordinary zeros, shifted by \(\tau=2\), begin at height \(12.13\) rather than \(14.13\), with the same gaps. This is a change of variables, not new arithmetic.

**[known]** A Dirichlet character is also a sequence of phases, but \(\chi(n)\) is periodic (or eventually periodic) in \(n\). There is no complex \(s_0\) with \(\chi(n)=n^{-s_0}\) for all \(n\). So \(L(s,\chi)\) is not a translate of \(\zeta\). For the primitive odd character \(\chi_{-4}\), a real Hardy function was formed from the usual gamma factor at conductor 4, and the first eight critical-line zeros were found numerically at heights \(6.02, 10.24, 12.99, 16.34, 18.29, 21.45, 23.28, 25.73\). The first of these is the standard first zero of \(L(s,\chi_{-4})\); it is not \(14.13\) minus anything natural.

A local-factor phase \(a_p=e^{i\theta}\) is a third, hybrid case: it rigidly shifts *one* column and leaves the inherited zeros alone.

**What “phase-like but not movement” means here.** Only the logarithmic phase \(n^{-i\tau}\) is a rigid motion of the whole zero set. A character is a phase of coefficients that changes the function, not a motion of the existing zeros.

## 3. Phase of the order zeta, and Gram’s law

**[known]** Ordinary \(\zeta\) has the real Hardy function \(Z(t)=e^{i\theta(t)}\zeta(\tfrac12+it)\), with Riemann–Siegel phase \(\theta(t)=\arg\Gamma(\tfrac14+it/2)-(t/2)\log\pi\). Gram points are the solutions of \(\theta(g_n)=n\pi\). Gram’s law is the heuristic that \((g_n,g_{n+1})\) contains exactly one zero. It fails; the first failure is a well-known block around height 280.

**The half-line, [project].** Voight’s completion for these definite rational quaternion orders is

\[
\Lambda_D(s)=D^s\,2^{2-2s}\,\pi^{1-2s}\,\Gamma(2s)\,Z_D(s),\qquad \Lambda_D(s)=\Lambda_D(1-s).
\]

On the critical line of this functional equation, \(\Lambda_D(\tfrac12+it)\) is real (checked at eight heights, imaginary parts below \(10^{-27}\)). The corresponding phase and Hardy function are

\[
\theta_D(t)=t\log D-2t\log 2-2t\log\pi+\arg\Gamma(1+2it),\qquad
H_D(t)=e^{i\theta_D(t)}Z_D(\tfrac12+it).
\]

\(H_D\) is real, and its nonreal zeros are exactly the local zeros, unconditionally: on \(\operatorname{Re}s=\tfrac12\), the two zeta factors in \(Z_D\) sit on \(\operatorname{Re}=1\) and \(\operatorname{Re}=0\) and do not vanish at positive height. So the completed order zeta *does* have a Riemann–Siegel-type real normalisation on the half-line, and *not* on the quarter-line: the functional equation pairs \(\tfrac14+it\) with \(\tfrac34-it\), so \(\Lambda_D(\tfrac14+it)\) is not real in general.

The inherited zeros, on the quarter-line, are the zeros of the ordinary Hardy function at double height, \(Z(2t)\). Their natural Gram points are ordinary Gram points at half height. That law is ordinary Gram’s law, transported, not a new quaternionic law.

**Counts, [project].** Ordinary-zero brackets are the already certified list through height 2,000 (1,517 zeros). Gram points for \(\theta\) and for \(\theta_D\) were isolated in interval arithmetic, with widths below \(10^{-21}\), much smaller than the nearest-zero gaps (\(5.4\cdot 10^{-4}\) ordinary, \(6.7\cdot 10^{-6}\) local).

- Inherited / ordinary, through height 2,000: 1,358 of 1,516 intervals hold the law; 158 fail, evenly split as 79 empty and 79 with two or more, in alternating blocks. The first failure is the classical one: \((g_{125},g_{126})\approx(280.80,282.45)\) is empty, and the next interval contains the two zeros at 282.47 and 283.21. Through ordinary height 200 — the inherited window of height 100 — every interval holds.
- Local / order completion, through height 1,000: 925 of 4,115 intervals hold; 3,114 are empty; 76 contain two or more (two different ramified primes falling in one short Gram interval). The mismatch is by design: Stirling gives \(\theta_D(t)\sim 2t\log t\), so Gram points densify like \((2/\pi)\log t\), while local zeros have constant density \((\log 30)/\pi\).

There *is* an exact “Gram law” for a single local factor in isolation: after removing the unimodular factor, \(1-p^{-2it}=2i\,e^{-it\log p}\sin(t\log p)\), whose zeros sit exactly at the arithmetic progression \(\pi k/\log p\). That tautology is not the Gram law of the completed order zeta.

**What “phase-like but not movement” means here.** The Riemann–Siegel phase is a description of where a real function vanishes, not a motion of zeros. Gram’s law is a heuristic about that alignment. For inherited zeros it is ordinary Gram’s law in other units; for local zeros against the genuine completion, it mostly fails, for a density reason.

## 4. The quaternion angle

**[known]** If \(I\) is a unit imaginary quaternion, \(I^2=-1\) and \(e^{I\theta}=\cos\theta+I\sin\theta\). Conjugation \(I\mapsto e^{-J\theta/2}I e^{J\theta/2}\) rotates the sphere of imaginary units. Because the Riemann zeta has real coefficients, it is slice-invariant: a complex zero \(x+yi\) gives a whole two-sphere \(\{x+yI:I^2=-1\}\) of zeros in \(\mathbb{H}\). Changing the slice does not move a zero; it looks at another representative of the same sphere.

**[project]** The same distinction appeared in the Q8 Euler product (`frobenius-quaternions.md`): the reduced norm of the quaternion product is the Artin L-value, independently of order and of signs inside conjugacy classes, but the direction in \(\mathbb{H}\) moved by tens of degrees under those choices. The length is arithmetic; the phase is bookkeeping.

**What “phase-like but not movement” means here.** \(e^{I\theta}\) is a symmetry orbit, or a non-canonical direction. It is not a deformation parameter.

## 5. Figures

One relation per figure. Desktop and 360px mobile, SVG and PNG.

- `figures/zero-phase-dial.svg` (and `.png`, plus `-mobile`): \(r\) slides the column to a new line; a \(2\pi\) turn in \(\theta\) raises every zero one slot and leaves the set unchanged.
- `figures/zero-phase-twists.svg` (and `.png`, plus `-mobile`): the first eight zeros of \(\zeta\), of \(\zeta(s+2i)\), and of \(L(s,\chi_{-4})\). Only the middle row is a rigid motion.
- `figures/zero-phase-gram.svg` (and `.png`, plus `-mobile`): the quarter-line and the half-line through height 40, with zeros and Gram-type ticks, and the full-window counts.
- `figures/zero-phase-sphere.svg` (and `.png`, plus `-mobile`): a zero sphere as the phase orbit of a complex zero, not a path.

## 6. What is exact, certified, or numerical

**Exact.** The local-zero formula, the \(2\pi\) monodromy, the sideways role of \(r\); the identity \(n^{-i\tau}\leftrightarrow s\mapsto s+i\tau\); the reason a character is not such a shift; the form of the Voight completion and of \(\theta_D\); the identification of half-line zeros of \(Z_D\) with the local lattice; the transport of ordinary Gram’s law to the inherited family; the Stirling density of \(\theta_D\); the slice-sphere symmetry.

**Certified interval arithmetic [project].** Ordinary zeros through height 2,000, from the existing height certificate; isolation of ordinary and order-zeta Gram points, widths below \(10^{-21}\); the Gram-law counts and the nearest-zero gaps.

**Numerical [project].** Reality of \(\Lambda_D(\tfrac12+it)\) and of \(H_D(t)\) at sample heights, imaginary parts \(10^{-27}\) and smaller; the eight zeros of \(L(s,\chi_{-4})\), residuals about \(10^{-30}\), with a real Hardy function checked at four heights.

## 7. Reproducibility

From this directory:

```bash
/workspace/tools/pari-venv/bin/python zero_phase.py
python3 zero_phase_figures.py
python3 test_zero_phase.py
```

The Gram computation uses python-flint and the existing height certificates in `order-height/`. It does not write into that directory. The figure and test scripts use system Python. This run does not deploy, and it does not touch `zeta-zeros-build/`.

Artifacts: `zero-phase.md`; `zero_phase.py`, `zero_phase_figures.py`, `test_zero_phase.py`; `zero-phase-results.json` and `zero-phase/results.json`; the eight figure stems in `figures/`.


## Conversion framing (Grant, 2026-11-10)
[known] The explicit formula converts between primes and zeros in both directions (like a Fourier transform): each zero gamma adds a wave x^{1/2} e^{i gamma log x}, and the prime staircase is their sum. Grant's framing: a conversion, not a filter.
