# Computation stretch — 13 November 2026, ~10:12–10:50 PM PST

Notes only; nothing was deployed. Scripts and JSON live in
`/workspace/math/lit/foliations/compute/`. Figures in
`/workspace/math/lit/foliations/figures/`. Tools: cypari2 / PARI 2.17.2,
python-flint (Arb) at 384 bits, system Python for SVG path-text.

Labels: **[known]** published identities; **[project]** this stretch’s
enclosures and enumerations. No **[speculative]** claims. This is not
progress on the Riemann hypothesis.

---

## 1. Deninger’s circle toy

**[known]** Deninger 2008, “The simplest example”; also [D1] §2, Lerch.
The space is the circle \(X=\mathbb{R}/(\log q)\mathbb{Z}\), foliated by
points, with \(\mathbb{R}\) acting by translation. The infinitesimal
generator \(\theta=d/dx\) on \(C^\infty(X)\) has eigenvalues
\(2\pi i\,\nu/L\), \(L=\log q\), \(\nu\in\mathbb{Z}\). The Ruelle zeta of
the flow is \(\zeta_R(s)=(1-q^{-s})^{-1}\). Deninger claims

\[
\det_\infty(s\cdot\mathrm{id}-\theta)\;=\;1-q^{-s},
\]

and that the leading Laurent coefficient of \(\zeta_R\) at the simple
pole \(s=0\) is \(1/\log q\).

### 1.1 Regularized determinant — proof  [known]

Write \(\lambda_n=c(n+\alpha)\) for the eigenvalues of \(s\cdot\mathrm{id}-\theta\),
with \(c=-2\pi i/L\) and \(\alpha=i L s/(2\pi)\). The spectral zeta is
built from the Hurwitz zeta, taking the principal branch of the argument
for \(s>0\) real (so \(\alpha\) lies in the upper half-plane):

\[
g(w)
=\zeta_H(w,\alpha)+e^{-i\pi w}\zeta_H(w,1-\alpha),
\qquad
\sum_n\lambda_n^{-w}=c^{-w}\,g(w).
\]

The identity \(\zeta_H(0,a)=\tfrac12-a\) gives \(g(0)=0\), so the
constant \(c^{g(0)}\) drops out of the regularized determinant
\(\det_\infty=\exp(-g'(0))\). Differentiating and using the known
evaluation \(\zeta_H'(0,a)=\log\bigl(\Gamma(a)/\sqrt{2\pi}\bigr)\) yields

\[
g'(0)=\log\frac{\Gamma(\alpha)\Gamma(1-\alpha)}{2\pi}-i\pi\bigl(\alpha-\tfrac12\bigr).
\]

The reflection formula \(\Gamma(z)\Gamma(1-z)=\pi/\sin(\pi z)\) then
gives, after a short calculation with the same branch,

\[
\exp(-g'(0))=2\sin(\pi\alpha)\,e^{i\pi(\alpha-1/2)}=1-q^{-s}.
\]

The branches cancel for \(s>0\) real. The identity extends by analytic
continuation (both sides are entire functions of \(s\); the right-hand
side is visibly entire).

### 1.2 Arb enclosure  [project]

Script: `compute/circle_toy.py`. Results: `compute/circle-toy-results.json`.
Figure: `figures/circle-toy-det.svg`.

At 384 bits, on a 58-point grid with \(q\in\{2,3,5,10\}\) and \(s\) both
real and complex (including \(\tfrac12\pm i\), \(2+3i\), and small
positive \(s\)), every Arb ball for

\[
\det_\infty^{\text{(Gamma formula)}}(s,q)\;-\;(1-q^{-s})
\]

contains \(0\) (typical radius \(\sim 10^{-115}\); max radius
\(\sim 10^{-114}\)). In the 52 rows that record \(g(0)\), the ball for
\(g(0)\) contains \(0\).

A symmetric-difference check of \(g'(0)\) with step \(h=1/256\) was
coded but is **not** a certificate: the \(O(h^2)\) remainder is large at
that step, and no row was “close.” The proof is the closed form in §1.1;
the grid is an enclosure of that closed form against \(1-q^{-s}\).

### 1.3 Leading coefficient  [known], with samples [project]

Figure: `figures/leading-coefficient.svg`.

Elementary: \(1-q^{-s}=s\log q+O(s^2)\), so \(\zeta_R(s)=(1-q^{-s})^{-1}\)
has a simple pole at \(s=0\) of residue \(1/\log q\). Equivalently,
\(1-q^{-s}=s L\,\psi(s)\) with \(\psi(s)=\int_0^1\exp(-s L t)\,dt\) and
\(\psi(0)=1\), hence \(s\zeta_R(s)\to 1/L\).

Samples of \(s\zeta_R(s)-1/\log q\) at \(s=\pm 10^{-k}\) (\(k=1,\dots,7\))
are \(O(s)\) (leading remainder \(s/2\)) and correctly **do not**
contain \(0\). That is a remainder, not a discrepancy. Recorded values:

- \(1/\log 2 \approx 1.4426950408\ldots\)
- \(1/\log 3 \approx 0.9102392266\ldots\)
- \(1/\log 10 \approx 0.4342944819\ldots\)

### 1.4 Lichtenbaum-style cup-product determinant  [known]

On \((H^\bullet(S^1,\mathbb{R}), D=\psi\cup)\) one has
\(\mathrm{rk}\,H^0=\mathrm{rk}\,H^1=1\), torsion \(0\), and
\(\det(\text{cup})=\log q\). The order formula
\(\sum_i(-1)^i\,i\,\mathrm{rk}\,H^i(S^1,\mathbb{Z})=-1\) matches
\(\mathrm{ord}_{s=0}\zeta_R=-1\), and
\(\zeta_R^*(0)=1/\det=1/\log q\). This is an exact algebraic remark
(coded as `lichtenbaum_det_of_cup()` in the script), not a numerical
check.

---

## 2. Poincaré–Hopf index sums

**[known]** Hopf 1927: a continuous vector field on a compact oriented
manifold, with isolated zeros, has \(\sum\mathrm{index}=\chi(M)\). For a
nondegenerate zero in dimension 2 the index equals \(\mathrm{sign}(\det DY)\).
Independently, the index is the winding number of \(Y/|Y|\) on a small
positively oriented circle.

Script: `compute/poincare_hopf.py`. Results: `compute/poincare-hopf-results.json`.
Figure: `figures/poincare-hopf-surfaces.svg`.

### 2.1 Height gradient on \(S^2\)  [project] calculation of a [known] example

\(Y(p)=e_z-(p\cdot e_z)p=(-xz,-yz,1-z^2)\). The only zeros are the poles
(because \(Y=0\) forces \(x=y=0\) and \(z=\pm 1\)). In the graph chart
over the upper hemisphere the linearisation at \(N\) is \(-I\)
(attracting node, \(\mathrm{sign}\det=+1\)); over the lower hemisphere
the linearisation at \(S\) is \(+I\) (repelling node, \(\mathrm{sign}\det=+1\)).
Discrete winding numbers on 720-gons of radius \(0.05\) in these charts
are \(+1\) and \(+1\). Sum of indices \(=2=\chi(S^2)\).

### 2.2 Sine field on \(T^2=\mathbb{R}^2/\mathbb{Z}^2\)

\(Y=(\sin 2\pi x,\sin 2\pi y)\). Zeros exactly at
\(\{0,1/2\}\times\{0,1/2\}\). The Jacobian is
\(DY=2\pi\,\mathrm{diag}(\cos 2\pi x,\cos 2\pi y)\), so
\(\mathrm{sign}(\det DY)\) is \(+1,-1,-1,+1\) (source, saddle, saddle, sink).
Windings on 720-gons of radius \(0.1\) in the cover agree with these
four signs. Sum of indices \(=0=\chi(T^2)\).

The Kronecker field (constant, irrational slope) has no zeros, so the
empty sum is \(0=\chi(T^2)\) as well — consistent, and the reason a
nonsingular foliation of the torus is possible.

### 2.3 Morse function \(f=z^2+x\) on \(S^2\)

Critical points of \(f|_{S^2}\) at \(\varepsilon=1\), from
\(\nabla f=\lambda\nabla g\) with \(g=x^2+y^2+z^2-1\):

- \(P_\pm=\bigl(\tfrac12,0,\pm\sqrt{3}/2\bigr)\), Lagrange multiplier \(\lambda=1\)
- \(E_+=(1,0,0)\), \(\lambda=\tfrac12\)
- \(E_-=(-1,0,0)\), \(\lambda=-\tfrac12\)

The tangential Hessian is \(D^2f-2\lambda I\) restricted to \(T_p S^2\).
In the orthonormal frame \(\{(0,1,0),(-z,0,x)\}\) this is exact:

| point | \(\det H\) | \(\mathrm{tr}\,H\) | Morse index | PH index of \(+\nabla f\) |
|-------|-----------:|-------------------:|------------:|--------------------------:|
| \(P_\pm\) | \(3\) | \(-7/2\) | \(2\) (local max) | \(+1\) (sink) |
| \(E_+\) | \(-1\) | \(0\) | \(1\) (saddle) | \(-1\) |
| \(E_-\) | \(3\) | \(4\) | \(0\) (local min) | \(+1\) (source) |

(The JSON records `detH ≈ 2.9999999999999996` at \(P_\pm\), a floating
\(\sqrt{3}\) artefact; the value is exactly \(3\).) Sum of Poincaré–Hopf
indices \(=1+1-1+1=2=\chi(S^2)\).

All three examples passed (`T2_ok`, `S2_height_ok`, `S2_morse_ok` true).

---

## 3. Function-field / elliptic-curve-over-\(\mathbb{F}_q\) orbit bookkeeping

**[known]** This is the one case of Deninger’s dictionary that is
actually constructed (Deninger [D3], 2001/2008): closed points of a
curve over \(\mathbb{F}_q\) correspond to Frobenius orbits, with

\[
\ell(\gamma)=|o|\log q, \qquad N\gamma=q^{|o|}=Nx.
\]

We did **not** construct a solenoid. We checked the orbit–zeta
bookkeeping against the Hasse–Weil zeta of an elliptic curve. The
Riemann hypothesis for curves over finite fields is a theorem (Weil);
we use only the known shape of \(Z(E,T)\) and the Hasse bound
\(|a|\le 2\sqrt{q}\). No RH claim for \(\zeta(s)\).

Script: `compute/elliptic_orbits.py`. Results: `compute/elliptic-orbits-results.json`.
Figure: `figures/elliptic-orbits.svg`.

### 3.1 What was checked  [project], against [known] identities

For an elliptic curve \(E/\mathbb{F}_q\) with trace \(a=q+1-\#E(\mathbb{F}_q)\),

\[
Z(E,T)=\frac{1-aT+qT^2}{(1-T)(1-qT)},
\qquad
\zeta_E(s)=Z(E,q^{-s})=\prod_x(1-Nx^{-s})^{-1}.
\]

Point counts \(N_n=\#E(\mathbb{F}_{q^n})\) come from the exact recurrence
\(N_n=q^n+1-t_n\), \(t_n=a\,t_{n-1}-q\,t_{n-2}\), \(t_0=2\), \(t_1=a\),
with \(a\) from PARI `ellap`. Three-way agreement, where available:

1. the recurrence (exact integers);
2. PARI `ellcard` of `ellinit(E, ffgen(F_{q^n}))` for every \(n\le 16\)
   (and \(n\le 24\) on one curve);
3. a naive enumeration of \(x\in\mathbb{F}_{q^n}\), counting solutions in
   \(y\) by the Euler / Legendre criterion on the Weierstrass
   discriminant, plus the point at infinity, for \(n\le 3\) (and
   \(n\le 4\) on \(y^2=x^3+1\) over \(\mathbb{F}_5\)).

(The integer prime power \(q^n\) must **not** be passed as the field
argument to `ellinit`; that is the wrong API and returns a wrong count.
The finite-field generator is the correct construction.)

Then \(P_n=\frac1n\sum_{d\mid n}\mu(d)\,N_{n/d}\) (an integer, and
non-negative) is the number of closed points of degree \(n\), i.e. the
number of Frobenius orbits of length \(|o|=n\), and
\(\sum_{d\mid n}d\,P_d=N_n\).

The Euler product identity

\[
\prod_{n\ge 1}(1-T^n)^{-P_n}
=\frac{1-aT+qT^2}{(1-T)(1-qT)}
\]

was checked as a formal power series with exact integer coefficients,
up to \(T^{16}\) on every curve and up to \(T^{24}\) on
\(y^2=x^3+1/\mathbb{F}_5\).

The leading Laurent coefficient of \(\zeta_E(s)\) at \(s=0\) is
elementary from the simple pole of \(Z(E,T)\) at \(T=1\):

\[
\lim_{s\to 0}s\,\zeta_E(s)=\frac{N_1}{(1-q)\log q}.
\]

### 3.2 Curves  (all agreed)

Eleven curve / field pairs, all of good reduction, plus a depth-24 run
on the first curve. `summary.all_ok` is true.

| curve | \(q\) | \(a_q\) | \(N_1\) | \(P_1\) | series | naive | PARI \(\mathbb{F}_{q^n}\) |
|-------|------:|--------:|--------:|--------:|:------:|:-----:|:------------------------:|
| \(y^2=x^3+1\) | 5 | 0 | 6 | 6 | yes | yes | 16 of 16 |
| \(y^2=x^3+1\) | 7 | \(-4\) | 12 | 12 | yes | yes | 16 of 16 |
| \(y^2=x^3+1\) | 13 | 2 | 12 | 12 | yes | yes | 16 of 16 |
| \(y^2=x^3-x\) (CM) | 5 | \(-2\) | 8 | 8 | yes | yes | 16 of 16 |
| \(y^2=x^3-x\) | 7 | 0 | 8 | 8 | yes | yes | 16 of 16 |
| \(y^2=x^3-x\) | 17 | 2 | 16 | 16 | yes | yes | 16 of 16 |
| 11a1 \(y^2+y=x^3-x^2\) | 7 | \(-2\) | 10 | 10 | yes | yes | 16 of 16 |
| 11a1 \(y^2+y=x^3-x^2\) | 13 | 4 | 10 | 10 | yes | yes | 16 of 16 |
| \(y^2=x^3+x+1\) | 5 | \(-3\) | 9 | 9 | yes | yes | 16 of 16 |
| \(y^2=x^3+x+1\) | 11 | \(-2\) | 14 | 14 | yes | yes | 16 of 16 |
| \(y^2=x^3+1\) (depth 24) | 5 | 0 | 6 | 6 | yes to \(T^{24}\) | \(n\le 4\) | 24 of 24 |

Example, \(y^2=x^3+1\) over \(\mathbb{F}_5\) (\(a=0\)):
\(P_n\) for \(n=1,\dots,8\) is
\(6,15,40,135,624,2620,11160,48600\). Orbit length \(|o|=n\),
\(\ell(\gamma)=n\log 5\), \(N\gamma=5^n\). Hasse bound \(a^2\le 4q\)
held on every curve.

---

## 4. Figure paths

All under `/workspace/math/lit/foliations/figures/`, dark-mode-safe
site palette, DejaVu path-text (rare glyphs substituted):

- `circle-toy-det.svg` — \(1-2^{-s}\) with sample points; enclosure summary.
- `leading-coefficient.svg` — \(s/(1-q^{-s})\) approaching \(1/\log q\) for \(q=2,3,10\).
- `poincare-hopf-surfaces.svg` — height field on \(S^2\); sine field on \(T^2\); Morse \(z^2+x\) on \(S^2\).
- `elliptic-orbits.svg` — \(\log P_n\) bars and orbit table for \(y^2=x^3+1/\mathbb{F}_5\); identity recap.

Earlier primer figures (unchanged this stretch):
`kronecker-foliation.svg`, `phase-portraits.svg`, `reeb-foliation.svg`,
`deninger-dictionary.svg`.

---

## 5. What this does not do

- It does not construct Deninger’s space \((X,\mathcal{F},\phi^t)\) for
  \(\mathrm{Spec}\,\mathbb{Z}\), nor a solenoid for these elliptic curves.
- It does not prove, or claim evidence for, the Riemann hypothesis.
- The finite-difference check of \(g'(0)\) is not a proof.
- Poincaré–Hopf is illustrated on three model fields, not a classification
  of singularities.
