# Local and inherited zeros: movement and collisions

10 November 2026, America/Los_Angeles. Question: do zeros move between the local and inherited families, and how do families of zeros behave under deformation?

(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 model family studied here; not a new number-theory theorem. [speculative] = an unproved additional idea.

This note does not claim RH progress, and it does not introduce a new joint field. The actual quaternion-order zeta factors, so its zeros do not interact. Interaction is studied in deliberately chosen model families, labelled as such.

## 0. Verdict

- **[known]** The ownership tree for these order zetas is the factorisation of Voight's formula. Each nonreal zero belongs to exactly one factor. In a single fixed function, nothing moves.
- **[known]** Replacing a local factor by \(1-a_p\,p^{1-2s}\) slides that prime's column rigidly. For \(a_p>0\) it sits at \(\operatorname{Re}s=\tfrac12+\log(a_p)/(2\log p)\). It lands on the quarter line at \(a_p=p^{-1/2}\) and on the three-quarter line at \(a_p=p^{1/2}\). Heights do not change.
- **[project]** Through heights 100, 1,000 and 10,000, no inherited height \(\gamma/2\) equals a local height \(\pi k/\log p\). The closest certified misses for \(D=30\) are \(0.019817348719\) at height 100, \(0.000206268854\) at 1,000, and \(0.000009241204\) at 10,000. Real dials therefore never produce a collision in these windows.
- **[known]** The factors multiply, so even an exact hit is only a double zero: the zeros pass through one another. There is no repulsion in this family. A complex dial can force such a hit; a numerical check at the first quarter-line zero confirms a double zero that splits again when the radius is nudged.
- **[known]** Scaling the argument, the family \(\zeta(\lambda s)\zeta(\lambda s-(\lambda-1))\), moves the whole inherited set together. On RH the two lines are \(\operatorname{Re}s=1/(2\lambda)\) and \(1-1/(2\lambda)\). This is the phrasal motion.
- **[project]** In interacting model families, neighbouring zeros on the critical line can collide and leave as the mirror pair \(s\) and \(1-\overline{s}\).

## 1. The ownership tree

**The order-zeta tree [known].** Voight, *Quaternion Algebras*, Theorem 26.4.9, equation 26.4.10, in the ideal-index coordinate:

\[
Z_D(s)=\zeta(2s)\,\zeta(2s-1)\prod_{p\mid D}(1-p^{1-2s}).
\]

For the definite rational algebras used here, \(D=2,3,30\) with ramified primes \(\{2\}\), \(\{3\}\), and \(\{2,3,5\}\). Writing \(Z_{\mathrm{split}}(s)=\zeta(2s)\zeta(2s-1)\) for the unramified product, one has \(Z_2=Z_{\mathrm{split}}\cdot(1-2^{1-2s})\) and \(Z_{30}=Z_{\mathrm{split}}\cdot\prod_{p\mid 30}(1-p^{1-2s})\), and so on. This is a ramification tree of orders, not a tower of number fields.

Ownership is exact, from the factorisation, and was already recorded in `order-off-axis.md` and `order-height.md`:

- \(\zeta(2s)\) owns the inherited zeros \(s=\rho/2\).
- \(\zeta(2s-1)\) owns the inherited zeros \(s=(\rho+1)/2\).
- Each local factor \(1-p^{1-2s}\) owns the column \(s=\tfrac12+i\pi k/\log p\), \(k\neq 0\).

On RH the first family lies on the quarter line and the second on the three-quarter line. That location, at all heights, is equivalent to ordinary RH and is not a new quaternionic statement. Distinct local columns cannot coincide, by unique factorisation. Complementary factors are nonzero at these nonreal points by the Euler product and the functional equation. At the centre the order is \(r-1\) if \(r\) primes ramify, so \(D=30\) has a double central zero.

In any one fixed \(Z_D\), nothing moves. Every zero sits where its owner puts it.

**The Q8 field tree is a different tree [known].** For the quaternion-group number fields, Heilbronn characters and Stark's simple-zero lemma describe ownership of zeros of Dedekind zetas among Artin factors; that was certified at the first three ordinary zeros and at \(\tfrac12\) in the downward-transfer work. It is not the ownership relation for these order zetas, and it is not used below.

## 2. The multiplicative dial

**[known]** Replace each local factor by \(1-a_p\,p^{1-2s}\), with a dial \(a_p\neq 0\). The zeros of that factor are

\[
s=\frac12+\frac{\operatorname{Log}a_p}{2\log p}-\frac{i\pi k}{\log p},\qquad k\in\mathbb{Z}\setminus\{0\}.
\]

For \(a_p>0\) the column is vertical at \(\operatorname{Re}s=\tfrac12+\log(a_p)/(2\log p)\), and the heights are unchanged: the column moves rigidly sideways. It lands on the quarter line at \(a_p=p^{-1/2}\) and on the three-quarter line at \(a_p=p^{1/2}\). The original middle line is \(a_p=1\). If \(a_p=re^{i\theta}\) with \(r>0\), the argument shifts the column vertically by \(\theta/(2\log p)\).

A collision with an inherited zero, for a real dial, needs two things at once: the same height, \(\gamma/2=\pi k/\log p\), and the column sitting on the inherited line, so \(a_p=p^{-1/2}\) or \(p^{1/2}\) (on RH). The height test does not depend on the real dial.

**[project]** Using the already certified ordinary-zero brackets from the order-height work, interval arithmetic gives a strictly positive minimum of \(\lvert\gamma/2-\pi k/\log p\rvert\) in each requested window. The \(D=30\) record (the closest of \(p=2,3,5\)) is:

| Height \(T\) | Lower bound | Upper bound | Witness |
|---:|---:|---:|---|
| 100 | 0.019817348719 | 0.019817348723 | ordinary zero 32 against the 27th \(p=5\) local height, near \(t=52.723\) |
| 1,000 | 0.000206268854 | 0.000206268858 | \(p=5\) |
| 10,000 | 0.000009241204 | 0.000009241208 | ordinary zero 18961 against the 4412th \(p=5\) local height, near \(t=8612.141\) |

The height-100 and height-1,000 bounds were recomputed here from the verified brackets. The height-10,000 bound is the already written certificate `order-height/inheritance-gap-10000.json`, cited, not rewritten. Each recorded closest pair is unique in its window. There is no exact height coincidence, so a real dial never produces a double zero through these heights.

The complementary factors remain the same as in the original classification, and they do not vanish at these points.

## 3. Passing through, and a forced hit

**[known]** \(Z(s;a)\) is a product. Zeros of different factors do not see one another. If a local zero and an inherited zero occupy the same point, the product has a double zero there (each factor simple, derivative of the product zero by the product rule). As the dial continues they separate into two simple zeros again. They pass through; they do not repel, and they do not change kind.

**[project]** A complex dial can arrange the hit. At the first ordinary zero, take \(s=\rho/2\) and \(a=2^{2s-1}\). This places a \(p=2\) local zero exactly on that inherited zero, with \(\lvert a\rvert=2^{-1/2}\), so the column is on the quarter line and the height has been shifted to match. High-precision evaluation of the \(D=30\) product and of its derivative are both tiny (about \(10^{-121}\) and \(10^{-60}\) at 60 digits). Scaling the modulus of \(a\) by \(0.98\) or \(1.02\) moves the local zero off the quarter line by about \(0.014\), leaving two nearby simple zeros at the same height. This is a numerical check of the product identity, not an interval certificate of the double zero.

## 4. Phrasal motion of the inherited set

**[known]** For a real scale \(\lambda>0\),

\[
\zeta(\lambda s)=0\iff s=\frac{\rho}{\lambda},\qquad
\zeta\bigl(\lambda s-(\lambda-1)\bigr)=0\iff s=\frac{\rho+\lambda-1}{\lambda}=\frac{\rho}{\lambda}+1-\frac1\lambda.
\]

The first family is the ordinary zero set scaled toward the origin by \(1/\lambda\): every inherited zero of this factor moves together. On RH the two lines are \(\operatorname{Re}s=1/(2\lambda)\) and \(\operatorname{Re}s=1-1/(2\lambda)\), mirrors of one another under \(s\mapsto 1-\overline{s}\). The quaternion-order inherited part is the case \(\lambda=2\), recovering the quarter and three-quarter lines at heights \(\gamma/2\).

So the inherited zeros of a factor move together, while each local column belongs to one prime and moves only as a block.

## 5. Interacting deformations

The genuine order zeta cannot supply an interaction: it is a product. Two model families, in the coordinate \(z=s-\tfrac12\), do.

**[project]** The even quartic \(P_\varepsilon(z)=(z^2+a^2)(z^2+b^2)+\varepsilon\), with \(a=7\), \(b=12\). For \(\varepsilon=0\) the zeros lie on the critical line at heights \(7\) and \(12\) (and their conjugates). The discriminant in \(w=z^2\) vanishes at the exact value \(\varepsilon=(a^2-b^2)^2/4=2256.25\), where the two positive-height zeros meet at \(t_\star=\sqrt{(a^2+b^2)/2}\approx 9.823\). For larger \(\varepsilon\) they leave the line as the horizontal mirror pair \(s\) and \(1-\overline{s}\) at that height. They do not collide with their conjugates at \(-t\).

**[project]** The same idea with three positive-height zeros, \(Q_\varepsilon(z)=(z^2+7^2)(z^2+8^2)(z^2+16^2)+\varepsilon\). The critical values of \(\varepsilon\) are the values at which the cubic in \(w=z^2\) has a double root, at \(\varepsilon\approx 11225.84\) and \(\varepsilon\approx-1172581.84\). As \(\varepsilon\) increases through the first of these, the closest adjacent pair \(7,8\) collides and leaves, and the isolated zero at height \(16\) stays on the line. On the scanned grid this family never had two off-line rectangles at once, so a two-dimensional crossing could not occur.

A high-degree monic pencil between a finite inherited set and a finite local set was tried and discarded: in double precision the endpoint residuals were of size \(20\), so those paths were not evidence.

## 6. (Removed 11 Nov 2026)

A speculative non-mathematical comparison was removed here. The mathematical facts it used remain: mirror pairs s, 1−s̄ are symmetric about Re s = ½ (functional equation); on the line, zeros leave only after neighbours collide; and the functional equation does not force off-line quadruplet rectangles to be nested (explicit product of two quartics along (ε_A, ε_B) = (700t, 11400t) at t = 1: lower pair δ ≈ 1.551, t ≈ 7.239; higher pair δ ≈ 0.913, t ≈ 17.287).

## 7. Figures

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

- `figures/zero-movement-dial.svg` (and `.png`, plus `-mobile`): the real dial. The local column slides from the half line through the inherited quarter-line column without touching. Hollow dots are inherited (\(0_{\mathrm{inh}}\), our label); solid dots and tracks are local (\(0_{\mathrm{loc}}\), our label). The closest miss through height 30 is marked; the certified records through 100 and 10,000 are in the caption.
- `figures/zero-movement-leave.svg` (and `.png`, plus `-mobile`): the exact quartic. Neighbouring zeros on the line move toward each other, meet at height \(9.82\), and leave as \(s\) and \(1-\overline{s}\).

## 8. What is exact, and what is numerical

**Exact.** The Voight factorisation and the ownership statement; the closed-form location of a local zero as a function of \(a_p\); the landings on the quarter and three-quarter lines; the product rule (zeros of a product pass through); the \(\zeta(\lambda s)\) scaling formulae; the quartic collision value and height; the hexic critical values from a double root of the cubic in \(w=z^2\); the functional-equation symmetry of mirror pairs about Re s = ½; the explicit crossing snapshot of the independent product.

**Certified interval arithmetic [project].** The strictly positive height gaps at \(T=100\) and \(T=1,000\), from the already verified ordinary-zero brackets, and the cited height-10,000 certificate. These are finite-window statements, not a proof that a height collision never occurs.

**Numerical [project].** The forced complex-dial double zero, checked at 60 decimal digits, not enclosed by balls. The algebraic model paths are exact in principle and are drawn from closed form or a degree-six companion matrix.

## 9. Reproducibility

From this directory:

```bash
/workspace/tools/pari-venv/bin/python zero_movement.py
python3 zero_movement_interact.py
python3 zero_movement_figures.py
python3 test_zero_movement.py
```

The gap computation uses python-flint in the PARI virtualenv and the existing height certificates in `order-height/`. It does not write into that directory. The interaction and figure scripts use system Python (numpy, cairosvg, mpmath). This run does not deploy, and it does not touch `zeta-zeros-build/`.

Artifacts: `zero-movement.md`; `zero_movement.py`, `zero_movement_interact.py`, `zero_movement_figures.py`, `test_zero_movement.py`; `zero-movement-results.json`, `zero-movement-interact-results.json`; copies under `zero-movement/`; the six figure stems in `figures/`.
