# 2O central L-values: interval certificate

Routine sprint, 13 November 2026, ~9:25–10:15 AM PST. Follows `polyhedral-fields-20261112.md` (float PARI check of the 2O field) and `tetra-central-simplicity-20261112.md` (the 2T certificate). Scripts and outputs: `octa1113/`.

Labels: [known] established, with source; [project] computed here; [speculative] unproved. **No RH claim.** This is a finite interval check of two Artin L-values at s = ½. It finds, moves or excludes no zero off the line, and it is not evidence for or against the Riemann Hypothesis.

## 1. Objects

**[project, from `polyhedral-fields-20261112.md`]** K = the PARI galpol real Galois field of degree 48, Gal(K/ℚ) = 2O = GAP [48,28] (binary octahedral, *not* GL(2,3) = [48,29]). It is ramified only at 2, 5 and 17. Its two faithful 2-dimensional Artin characters (PARI `galoischartable` columns 5 and 4) are

- ρ (column 5): conductor N = 924 800 = 2⁷·5²·17², Γ-factor Γ_ℝ(s)² (K is totally real, so complex conjugation acts trivially), root number −1;
- ρ⊗ε (column 4): same conductor and Γ-factor, root number +1. Here ε is the sign character of S₄ = 2O/{±1}, which on this field is the Legendre symbol (·/5), conductor 5. Also ρ⊗ε is the Galois conjugate of ρ under √2 ↦ −√2.

No LMFDB label: the LMFDB search pages and API pagination are captcha-blocked from this host (see the 12 Nov note).

## 2. Results

**[project, computer-assisted, interval arithmetic (Arb/FLINT via python-flint 0.9.0)]**

| quantity | enclosure (224-bit run, c·n ≤ 130) | 160-bit run, c·n ≤ 120 | PARI float (not used as input) |
|---|---|---|---|
| root number of ρ, from θ(½)/(2θ(2)) | −1.0000000000000 ± 2.0·10⁻¹⁵ | −1.00000000 ± 3.9·10⁻¹⁰ | −1 |
| root number of ρ⊗ε | 1.00000000000000 ± 1.1·10⁻¹⁶ | 1.0000000000 ± 2.1·10⁻¹¹ | +1 |
| **L′(ρ, ½)** | **1.94965615008486 ± 4.6·10⁻¹⁵** | 1.949656150 ± 3.3·10⁻¹⁰ | 1.94965615008486302… |
| **L(ρ⊗ε, ½)** | **4.734975816857929 ± 5.1·10⁻¹⁶** | 4.7349758169 ± 7.7·10⁻¹¹ | 4.73497581685792869… |

So, as theorems conditional only on the standard inputs listed in §4:

1. **L(ρ, s) has a simple zero at s = ½.** The root number −1 forces Λ(½) = −Λ(½) = 0, and L′(ρ, ½) > 1.9496561500848 ≠ 0. The order of the central zero is exactly 1.
2. **L(ρ⊗ε, ½) ≠ 0**, indeed L(ρ⊗ε, ½) > 4.7349758168579.

Both runs agree with each other and with PARI's floating-point values. The 12 Nov float slopes (0.9397 → 0.99994 for ρ; ≈ 0 for ρ⊗ε) are now backed by enclosures.

Wall time: 26 s (160-bit) and 65 s (224-bit) for the certificate; 116 s for the PARI export. This was well inside the ~2 h budget, so no fallback test was needed.

## 3. Method

The 2T script `tetra1112/certify_lprime.py` encloses the Mellin integral by interval Riemann sums with monotone K₀ hulls. At conductor 163² that already gave radius ≈ 0.4 on a value ≈ 3. The hull width grows like 1/c with c = 2π/√N, and here √N ≈ 961.7 against 163, so that method would need about 6× more cells per digit. Instead, `octa1113/certify_octa.py` integrates **term by term exactly**:

- For Γ_ℝ(s)²: Λ(s) = N^{s/2}Γ_ℝ(s)²L(s) = ∫₀^∞ θ(y) y^{s−1} dy with θ(y) = 4 Σ aₙ K₀(cny), c = 2π/√N. [known: Mellin transform of K₀]
- Λ is entire (Artin conjecture for this ρ, see §4) and Λ(s) = W Λ(1−s), so Λ(s) = ∫₁^∞ θ(y)(y^{s−1} + W y^{−s}) dy. Hence
  - W = −1: Λ′(½) = 4 Σ aₙ G₁(cn), with G₁(z) = 2∫₁^∞ K₀(zy) y^{−1/2} log y dy = 2z^{−1/2}[H(z) − log z · F(z)];
  - W = +1: Λ(½) = 4 Σ aₙ G₀(cn), with G₀(z) = 2z^{−1/2}F(z);
  - F(z) = ∫_z^∞ K₀(x)x^{−1/2}dx and H(z) = ∫_z^∞ K₀(x)x^{−1/2}log x dx.
- F and H are computed at every z = cn (n ≤ M) by chaining `acb.integral` (rigorous Gauss–Legendre with error bounds) over [cn, c(n+1)], downward from z = c(M+1). The integrand is K₀(x)x^{−1/2}(1 + i log x): its real part gives F and its imaginary part gives H in one integral. Above z = c(M+1) we use K₀(x) ≤ √(π/2x) e^{−x}, which gives F-tail ≤ √(π/2) e^{−X₀}/X₀ and H-tail ≤ √(π/2) e^{−X₀}.
- n > M: |aₙ| ≤ d(n) ≤ n (Frobenius eigenvalues are roots of unity; ramified factors are 1), G₁(z) ≤ 2√(π/2z) e^{−z}/z², G₀(z) ≤ 2√(π/2z) e^{−z}/z. Tail bounds: 3.9·10⁻⁴⁹ and 2.5·10⁻⁵¹.
- L′(½) = Λ′(½)/(N^{1/4} π^{−1/2} Γ(¼)²), because L(½) = 0 kills the derivative of the prefactor. L(½) = Λ(½)/(N^{1/4} π^{−1/2} Γ(¼)²). The completion factor is 229.98481860891791764.
- **Root numbers are measured, not assumed.** θ(1/y) = W y θ(y) at y = 2 uses all 40 000 coefficients for θ(½) (c·40 000/2 ≈ 131), plus an explicit tail. The ratio θ(½)/(2θ(2)) is −1 ± 2·10⁻¹⁵ for ρ and +1 ± 1·10⁻¹⁶ for ρ⊗ε. For ρ, θ(1) is enclosed in [±7.4·10⁻¹⁵], as W = −1 requires. This identity also cross-checks the conductor, Γ-factor and every coefficient up to about n ≈ 20 000 at once.
- Sanity: F(c) = 3.49212049397, and F(0) − F(c) = 1.1554 matches 2√c(2 − log(c/2) − γ) from the small-x expansion of K₀. (The note string inside the 160-bit JSON has a typo, "1 − log"; it is fixed in the script and in the 224-bit JSON.)

**Coefficients.** a_p for unramified p are read from PARI `lfunartin`/`lfunan` and recognised exactly in ℤ[√2] (values 2, −2, 1, −1, 0, ±√2). From them aₙ is built multiplicatively with det = 1 (ρ ⊂ SU(2)). Ramified primes 2, 5, 17 have Euler factor 1 (`lfuneuler` = 1 for both). Audits, all with zero mismatches:

- the exact aₙ agree with PARI's aₙ for every n ≤ 40 000 (largest difference 0.0);
- a_n(ρ⊗ε) = (n/5)·a_n(ρ) for every n ≤ 40 000;
- for all 4 200 unramified p ≤ 40 000, the residue degree f from `idealprimedec` (= order of Frob_p) matches a_p: f = 1 → 2, 2 → −2, 3 → −1, 6 → 1, 4 → 0, 8 → ±√2;
- ε(p) from PARI equals (p/5), and every order-8 Frobenius has ε = −1, as it must, since order-8 elements map to 4-cycles in S₄.

The ±√2 sign on order-8 primes is the one input that the degree audit cannot see: it is PARI's choice of embedding. It is pinned down by the θ functional-equation check, where a wrong sign set breaks the identity.

## 4. Inputs and hypotheses

- [known] Artin L(ρ, s) for 2-dimensional ρ with projective image S₄ (octahedral) is automorphic, hence Λ is entire with the standard functional equation: Langlands (tetrahedral) and Tunnell 1981 (octahedral). 2O/{±1} = S₄.
- [known] Self-dual ρ (real character) ⇒ W ∈ {±1}. Here W is in any case measured to 10⁻¹⁵.
- [known] K₀(x) ≤ √(π/(2x)) e^{−x} for x > 0 (K_ν increases in ν ≥ 0, and K_{1/2}(x) = √(π/(2x)) e^{−x}); Mellin ∫K₀(x)x^{s−1}dx = 2^{s−2}Γ(s/2)².
- [project input] Conductor 924 800, Γ-factor and local factors at 2, 5, 17 come from PARI 2.17 `lfunartin`. They are not independently re-derived from ramification groups, but they are consistent with the θ identity to 10⁻¹⁵.
- Not claimed: anything about zeros off s = ½, any RH statement, or the 2I field, where the character field is ℚ(√5) and PARI's `galoisinit` fails.

## 5. What this does and does not settle

- [closed — project, computer-assisted, interval arithmetic] The 2O forced central zero of L(ρ) is simple, and L(ρ⊗ε, ½) ≠ 0. These are exactly the two L-value inputs that the 2T argument needed: there, the 3/8 exponent used the simple zero of L(ρ) and the non-vanishing of the twists at ½.
- [open — project] For 2T, the aligned product was written exactly as a product of fractional powers of L-functions (`certify_exponents.py`, the 3/8 identity). **No such 2O identity has been derived.** So these certificates do not by themselves give a 2O central exponent. The 12 Nov note's statement that "the predicted branch power at t = 0 is the order of the forced zero" stays a heuristic, labelled as such. The next step is the 2O decomposition of the aligned product, with the order-8 and order-4 classes, into L(ρ), L(ρ⊗ε), ζ, L(ε) and the 3-dim pieces, followed by the same bookkeeping as for 2T.
- [open] The same for 2I needs L-values over ℚ(√5). Without `galoisinit`, the coefficients would have to come from the cached automorphisms (poly1112).

## 6. Files

- `octa1113/export_pari.py`, `export.log`, `pari-data.json` (1.6 MB: aₙ to 40 000 for ρ, ρ⊗ε, ε; residue degrees; PARI floats)
- `octa1113/certify_octa.py`; runs `certify-cn120-p160.log` / `certify-octa-results-cn120-p160.json` and `certify-cn130-p224.log` / `certify-octa-results-cn130-p224.json` (the headline)
- Python: `/workspace/tools/pari-venv/bin/python` (cypari2, python-flint 0.9.0).
