# Interval certificates for the two icosahedral (2I) forced central zeros — 14 Nov 2026, ~12:25–12:45 PM PST

Labels: [known] established, with source; [project] computed here; [speculative] unproved. **No RH claim.** The zeros at s = ½ studied here are *forced* by the sign (root number −1) of the functional equation; this note only proves that each of them is **simple** (L′(½) ≠ 0). It finds, moves or excludes no zero off the line.

Follows `icosa-oddW-20261114.md` (this morning: both fields have W = −1, float + local proof). Scripts and outputs: `icosa-cert1114/`.

## 1. Result  [project, computer-assisted, interval arithmetic (Arb via python-flint 0.9.0); conditional on Artin holomorphy]

For each field both faithful 2-dimensional characters were certified: ρ (the orientation of `cert1113/ap-oriented-*.json`) and its Galois conjugate ρ′ (√5 → −√5). Which of the two is LMFDB's `.a.a` and which `.a.b` is not decided here; both are covered.

| Field (LMFDB) | conductor | Γ-factor | rep | W (theta ratio θ(½)/(2θ(2)), interval) | **L′(½)** (cutoff c·n ≤ 80, 400 bits) | check run (c·n ≤ 60, 300 bits) | PARI float |
|---|---|---|---|---|---|---|---|
| 2.3806401.120.a (p = 1951) | 1951² = 3 806 401 | Γ_ℝ(s)² | ρ | −1 ± 1·10⁻³³ | **14.6404198415908494271272808 ± 4.9·10⁻²⁶** | 14.6404198415908494 ± 4.7·10⁻¹⁷ | 14.640419841590850183 |
| | | | ρ′ | −1 ± 4·10⁻³³ | **3.7050867341305652169151417 ± 5.2·10⁻²⁶** | 3.7050867341305652 ± 3.7·10⁻¹⁷ | 3.705086734130565115 |
| 2.4583881.120.a (p = 2141) | 2141² = 4 583 881 | Γ_ℝ(s+1)² | ρ | −1 ± 3·10⁻³⁸ | **8.38821569820444695438668613 ± 3.7·10⁻²⁷** | 8.388215698204446954 ± 7.4·10⁻¹⁹ | 8.388215698204446979 |
| | | | ρ′ | −1 ± 1·10⁻³⁶ | **6.188781635851670992970410311 ± 8.9·10⁻²⁸** | 6.188781635851670993 ± 3.8·10⁻¹⁹ | 6.188781635851670676 |

(The PARI floats lie within 8·10⁻¹⁶ of the intervals, PARI's float accuracy here.) All four intervals exclude 0, so — under the hypothesis below — **each of the four L-functions has a simple zero at s = ½**, forced by W = −1 and of order exactly 1. The two runs (different cutoff, precision and table length: M = 18 631 / 24 841 for 1951, 20 446 / 27 261 for 2141) agree inside their radii.

New this run: the conjugates ρ′ (also W = −1, as the local argument of the morning note predicts — it only uses central elements, which ρ and ρ′ treat alike) with L′(ρ′, ½) ≈ 3.705 and 6.189 [project]; PARI `lfuncheckfeq` for ρ′: 2⁻⁵⁰ (1951) and 2⁻⁴⁶ (2141) with W = −1; W = +1 fails (+1 / 0).

## 2. Hypotheses and what is and is not rigorous

- **Hypothesis H (Artin holomorphy):** Λ(s) = N^{s/2} Γ_ℝ(s+μ)² L(ρ, s) is entire. The functional equation Λ(s) = W Λ(1−s) holds unconditionally (Brauer induction) [known], and W = ±1 because ρ is self-dual with trivial determinant: W(ρ)W(ρ*) = det ρ(−1) = 1 and W(ρ*) = conj W(ρ) [known: Tate, Corvallis 1979]. (ρ is symplectic, LMFDB indicator −1, so Fröhlich–Queyrut, which forces W = +1 for orthogonal ρ, does not apply [known].) Holomorphy is Artin's conjecture, open for these reps: 1951 is an even icosahedral rep [known: open; cf. Booker, Experiment. Math. 15 (2006)]; 2141 has ρ(c) = −I ("totally odd", det(ρ(c)) = +1), so Khare–Wintenberger (which needs ρ odd, det ρ(c) = −1) does not apply [known].
- Under H: the split-at-1 formula below, the theta relation θ(1/y) = W y θ(y), and hence **the root number −1 itself is certified by interval arithmetic** (ratio −1 ± 10⁻³³, which excludes +1). Without H the morning's local proof of W = −1 (Tate's ε(χ)ε(χ⁻¹) = χ(−1) at abelian decomposition groups) still stands [project, proved from known local facts]; the interval ratio is then an extra, independent confirmation.
- **Coefficients:** the a_p come from the oriented-discriminant Frobenius classification of `cert1113/oriented_2I.py` (PARI computations, not themselves interval-certified). Audits: (i) the exact ℤ[φ] coefficients agree with the float vector that PARI's `lfuncheckfeq` accepted to 2⁻⁵⁵ / 2⁻⁵⁴ (max diff 9·10⁻¹⁶, all n ≤ 119 999); (ii) independent audit against the LMFDB projective quintic (`Proj_Polynomial`): the quintic's cycle type mod p (A₅ image order 1 / 2 / 3 / 5) matches |a_p| ∈ {2} / {0} / {1} / {φ, φ−1} at all 2 258 unramified primes p < 20 000, 0 mismatches, both fields (`pari_cross.py`); (iii) the interval theta ratio = −1 to 33 digits is a further global check of the whole coefficient set. The sign inside each ±-pair (5A vs 5B orientation, and ±1 lifts) is not audited independently of the morning's method beyond (i) and (iii). At p = 2 the oriented test is blind; for 2141 a₂ is fixed by the functional equation (morning note §1), for 1951 Frob₂ has order 3/6 (unaffected).
- **Ramified Euler factor = 1** [project, from `cert1113/ram_check.py`]: inertia at 1951 is C₁₀ ∋ −1 (acts as −I), at 2141 it is C₅ (eigenvalues ζ₅^{±1} ≠ 1); in both cases V^I = 0, consistent with conductor exponent 2 = dim V − dim V^I (tame).
- **Tail bounds** use |a_n| ≤ d(n) ≤ n (checked for every n ≤ 119 999, 0 violations) and K₀(x) ≤ √(π/(2x)) e^{−x} [known, standard].

## 3. Method  [project; same scheme as `octa1113/certify_octa.py`, `icosa-exp1113/certify_icosa.py`]

φ_μ(x) = 4 x^μ K₀(2πx) has Mellin transform Γ_ℝ(s+μ)² (Mellin of K₀: ∫₀^∞ K₀(x) x^{s−1} dx = 2^{s−2} Γ(s/2)² [known, DLMF 10.43.19]; checked numerically at s = 2 to 30 digits in the script). With θ(t) = Σ a_n φ_μ(n t/√N), Λ(s) = ∫₀^∞ θ(t) t^{s−1} dt; under H, θ(1/t) = W t θ(t), and splitting at t = 1 with W = −1:

  Λ′(½) = 2 Σ a_n ∫₁^∞ φ_μ(n t/√N) t^{−1/2} log t dt = P_μ Σ a_n z^{−1/2} [H_μ(z) − log z · F_μ(z)],  z = 2πn/√N,

with P₀ = 8, P₁ = 4/π, F_μ(z) = ∫_z^∞ K₀(x) x^{μ−1/2} dx, H_μ(z) = ∫_z^∞ K₀(x) x^{μ−1/2} log x dx (Γ_ℝ(s)² → K₀ kernel; Γ_ℝ(s+1)² → x·K₀ kernel — note it is x K₀(x), not e^{−x}; e^{−x} belongs to Γ_ℝ(s)Γ_ℝ(s+1) = Γ_ℂ(s)). Then L′(½) = Λ′(½) / (N^{1/4} Γ_ℝ(½+μ)²). F, H are tabulated by chaining rigorous `acb.integral` over [z_n, z_{n+1}] downward from z_{M+1}, with explicit tails F ≤ √(π/2) e^{−X}/X^{1−μ}, |H| ≤ √(π/2) e^{−X}(log X + 1)/X^{1−μ}; the n > M tail is bounded by √(π/2) c^{μ−5/2} Σ_{n>M} n^{1/2} e^{−cn} (ratio-test geometric bound). Coefficients are exact in ℤ[φ], φ = (1+√5)/2, built multiplicatively with a_{p^{k+1}} = a_p a_{p^k} − a_{p^{k−1}} (det ρ = 1) and fed to Arb with exact √5.

**Lesson (precision):** the first test (cutoff 40 at 96 bits) gave radii ~10⁻⁸ because `bessel_k` of an acb argument loses ≈ 2x·log₂e bits for x around the cutoff (cancellation in the small-argument series), so the chained pieces near the top were wide. Working precision must exceed ≈ 2.9·cutoff + target bits; hence 300 bits for cutoff 60 and 400 bits for 80. The same effect explains the ~10⁻¹⁰ radii of the 13 Nov 2O/2I runs (cutoff 120 at 160 bits) — those enclosures are still valid, just wider than necessary [project].

Runtimes: 80–200 s per field and run on one core.

## 4. Figure

`figures/icosahedral-odd-sign-site.svg` (+ `-mobile.svg`, PNG previews): Hardy-type real function Z(t) on the critical line, |t| ≤ 3, for ρ and ρ′ in both fields (PARI `lfunhardy`, float; each curve scaled to its own maximum), with the forced crossing at t = 0 marked and the four interval values. Data `icosa-cert1114/hardy-data.json`; script `icosa-cert1114/site_figure.py`. Next zeros above the centre (float): 1951 ρ 1.183, ρ′ 0.586; 2141 ρ 0.503, ρ′ 0.618 [project, float, uncertified].

## 5. Files

- `icosa-cert1114/certify_oddW.py` — the certificate (args: label, cutoff, bits); results `certify-oddW-results-<label>-cn{60,80}-p{300,400}.json`, logs `run-*.log`.
- `icosa-cert1114/pari_cross.py` → `pari-cross-results.json`, `pari-cross.log` — PARI floats for ρ and ρ′, quintic audit.
- `icosa-cert1114/lmfdb-artin-*.json` — LMFDB API records (artin_reps), fetched 14 Nov 2026.
- `icosa-cert1114/hardy_data.py`, `hardy-data.json`, `site_figure.py`; `figures/icosahedral-odd-sign-site*.{svg,png}`.

## 6. Still open
- Aligned-product central exponent for these two fields (both ρ and ρ′ have a simple forced zero — now certified under H — plus Sym³ρ to sort out).
- Unconditional handle on ord_{½} L(ρ) without Artin's conjecture.
- p = 2 blind spot of the oriented test (lift to ℤ₂ mod 8); an independent audit of the sign inside each ±-pair of a_p.

## 7. Page
Folded at ~12:45 PM PST as /critical-line-hypercomplex#binary-icosahedral-odd-sign (after #binary-icosahedral-exponent, in the same disclosure), with figure #hyper-binary-icosahedral-odd-sign; module `zeta-zeros-build/icosa_oddW_page.py`; deploy version bd8aba3e-6ac3-431c-893c-328b20e2274a, verified live. (The research copy of open-questions.md on the site predates its final "closed — site" line; it refreshes on the next build.)
