# A₅ central values L(3, ½), L(3′, ½), L(4, ½) for three 2I fields — 15 Nov 2026, ~4:45–5:30 PM PST (routine sprint, hypercomplex-critical-line-sprint)

Closes the open item "A₅ central values L(3,½), L(3′,½)" (open list since 13 Nov) for the galpol 2I field (conductor 2⁸·11²·23²) and the two root-number −1 fields 2.3806401.120.a (1951²) and 2.4583881.120.a (2141²). Scripts and data: `a5central1115/`. Figure: `figures/a5-central-values.png` (+ `.svg`).

Labels: **[known]** published, with source; **[project]** computed or proved here; **[speculative]** unproved. All numbers are floating point (PARI at 64 or 160 bits; an independent double-precision sum), **not** interval enclosures. Every value presupposes that the completed L-function is entire with the Artin functional equation (Artin's conjecture for these even icosahedral A₅ representations, **not known** [known: Booker, Experiment. Math. 15 (2006)]). Nothing here is about zeros of ζ or the Riemann Hypothesis.

## 0. Why this angle

Candidate 1 of the task. It was the cheapest item left that changes a stated result: the 1/24 theorem in `icosa-exponent-20261113.md` §6 and `icosa-oddW-aligned-exponent-20261114.md` §5 ends with "provided L(3, ½), L(3′, ½), L(4, ½) are nonzero (not computed)". The galpol data (Euler factors, conductors) were already fixed in `icosa-pure-zero-side-20261115.md`. For 1951/2141 the 13 Nov note said the oriented classes "only reach p < 20 000"; in fact `cert1113/ap-oriented-*.json` already holds oriented traces for **all** p < 1.2·10⁵, and the degree-3 L-functions of conductor p₀² need only 2.3–9.1·10⁴ coefficients (PARI `lfuncost` at 64 / 160 bits). So no new Frobenius computation was needed. Candidate 2 (701 family) needs 3·10⁸ coefficients at 2¹²; candidate 3 items need new theory.

## 1. Result

**[project, floats; conditional on Artin holomorphy]** All nine central values are nonzero and positive:

| field | L(3, ½) = L(Sym²ρ, ½) | L(3′, ½) = L(Sym²ρ′, ½) | L(4, ½) = L(ρ⊗ρ′, ½) |
|---|---:|---:|---:|
| galpol, 2⁸·11²·23² | 2.66438335895444219585 | 1.00195320417468542863 | 0.90634289558211999606 |
| 2.3806401.120.a, 1951² | 1.85829235550052917515 | 0.71500243731102316900 | 3.35362693157 (14 Nov) |
| 2.4583881.120.a, 2141² | 0.71955791719839336942 | 3.12566338689618199872 | 0.934617074359 (14 Nov) |

(20 digits shown from the 160-bit runs; the 14 Nov L(4) values are double-precision theta sums from `sym3-1114/`.) All have root number +1 (as they must: they are orthogonal, [known: Fröhlich–Queyrut 1973]), and `lfunorderzero` = 0.

**Consequence [project; conditional on Artin holomorphy for ρ, ρ′, Sym³ρ and the A₅ pieces; float nonvanishing of L(Sym³ρ,½) (1951/2141) or L′(Sym³ρ,½) (galpol)].** The proviso in the 1/24 theorem is now met for all three 2I fields, so

    F₁₀(s) ∼ c · (s − ½)^{1/24 + i(s₁+s₂)/10},  c ≠ 0, ∞,  s₁ = sin 36°, s₂ = sin 72°,

with the imaginary part of the exponent coming only from the pole of ζ(2s). The A₅ factors at s (weights i(s₂−s₁)/5, 2i s₁/5, −i(s₁+s₂)/5 on L(3), L(3′), L(4)) are entire, real and positive on [½, 2] (sampled every 0.01, no sign change, `real_segment.py`), so they contribute only a constant phase to c: φ_A₅ = Σ w_χ log L(χ, ½) = +0.1019 (galpol), −0.4063 (1951), +0.2648 (2141) rad (`phase.py`; orientation ρ ↔ trace 2cos 72° = 0.618 on the "A" order-5 class; swapping the orientation would give 0.2608, −0.2511, 0.0262).

## 2. Method and checks

**galpol field** (`galpol_pari.py` floats, `galpol_exact.py` exact). Coefficients from the Frobenius classes of `icosa-exp1113/frob/` via `icosa-zero1115/an_build.py` / `an_exact.py`; Euler factors from `icosa-pure-zero-side-20261115.md` §3: at 11, 23: (1+p^{−s})^{−1} for 3, 3′, (1−p^{−2s})^{−1} for 4; at 2: 1 for 3, 3′, (1−2^{−s})^{−1} for 4. Γ_ℝ(s)^d (totally real).
- 64-bit, 4·10⁵ float coefficients: checkfeq 2⁻⁵⁴ / 2⁻⁵⁵ / 2⁻⁵⁴ with W = +1, 0 or +1 with W = −1.
- 160-bit, exact ℤ[φ] coefficients (3·10⁵ for 3, 3′; 4.5·10⁵ for 4; PARI cost 1.7·10⁵ / 3.4·10⁵): checkfeq **2⁻¹⁷⁷, 2⁻¹⁷⁴, 2⁻¹⁶⁸**. 64-bit and 160-bit values agree to ≤ 2.3·10⁻¹⁶ relative (double-precision coefficient level).
- **Independent check of L(4) [project]:** L(4, s) = ζ_{K₅}(s)/ζ(s) for the A₅ quintic K₅ = LMFDB 5.5.16386304.1, x⁵ − 2x⁴ − 20x³ + 40x² + 56x − 80 (fetched from the LMFDB API; PARI: disc 16 386 304 = 2⁸·11²·23², `polgalois` A₅, totally real). Coefficient audit: #roots(Q₅ mod p) − 1 = a_p(4) at all 2 259 unramified p ≤ 20 000, 0 mismatches. PARI's own Dedekind zeta (no Frobenius table at all) gives L(4, ½) = ζ_{K₅}(½)/ζ(½) = 0.906342895582119996059637556923300417501461794528, **equal to all 48 printed digits** to the exact-coefficient value (`quintic_check.py`, `quintic-check-b160.json`). ζ_{K₅}(½) = −1.3236, ζ(½) = −1.4604.

**1951 / 2141 fields** (`oddW_sym2.py`). Oriented traces a_p(ρ) for all p < 1.2·10⁵ from `cert1113/oriented_2I.py`; Sym² trace a_p² − 1, ρ′ by σ(φ) = 1 − φ. Local data **[project, from known local facts]**: D at p₀ is abelian (`icosa-oddW-20261114.md` §4), Sym²ρ|_D = ψ² ⊕ 1 ⊕ ψ⁻², ψ² ramified (order 5 on inertia in both fields), so conductor p₀², Euler factor (1 − p₀^{−s})^{−1}; ρ(c) = ±1 acts trivially on Sym², so Γ_ℝ(s)³ for both fields; W_{p₀} = ψ²(−1)·1 = 1 (Tate ε(χ)ε(χ⁻¹) = χ(−1)), W_∞ = 1, so W = +1, matching the numerics.
- 64-bit (6·10⁴ coefficients, cost 2.3–2.6·10⁴) and 160-bit (1.2·10⁵ exact ℤ[φ] coefficients, cost 8.3–9.1·10⁴): checkfeq **2⁻¹⁷⁵, 2⁻¹⁷⁵, 2⁻¹⁷³, 2⁻¹⁷⁶** with W = +1; W = −1 gives 0 / +1; the wrong Γ-type [1,1,0] gives 2⁻³. The two runs agree to ≤ 1.3·10⁻¹⁸.
- L(2): PARI formula vs Euler product over p < 1.2·10⁵ (floats) agree to ≤ 5·10⁻⁹ (truncation-level, as expected).
- **Independent second method [project]** (`theta_indep.py`, no PARI L-function code): kernel φ(x) = 2 G^{3,0}_{0,3}(π³x² | 0,0,0) (mpmath Meijer G, inverse Mellin of Γ_ℝ(s)³), incomplete Mellin integrals by trapezoid on a log grid (h = 10⁻³), 4·10⁴ coefficients, cutoffs 2·10⁴ and 4·10⁴ identical. Agreement with PARI: relative 3.0·10⁻⁹, 3.1·10⁻⁹, 1.1·10⁻⁹, 1.2·10⁻⁹ (quadrature-limited; doubles). Theta relation θ(1/y) = y θ(y) at y = 1.1, 1.3, 1.7: 1 ± 3.5·10⁻⁸. (The Mellin normalisation check at s = 2 is off by 1·10⁻⁴ because the grid starts at x ≈ 2·10⁻⁴ and the missing piece ∫₀ φ x² dx/x ≈ 3·10⁻⁶ is never used by the sums.)
- **p = 2 audit for 2141 [project]:** the oriented test is blind at p = 2. With the flipped a₂ (−1.618 instead of 0.618) L(3) and L(3′) fail the functional equation (checkfeq 2⁻⁷ both, vs 2⁻¹⁷³ / 2⁻¹⁷⁶). So a second L-function (an A₅ one, where the ±1-lift ambiguity disappears) independently confirms the a₂ fixed on 14 Nov from L(ρ).

## 3. The tiny ways the quaternion structure shows up

1. **[project]** The √5-conjugate pair 3, 3′ (Sym² of ρ and of ρ′) are different L-functions with clearly different central values: ratios L(3,½)/L(3′,½) = 2.66, 2.60, 0.23. The orientation (which 5-cycle class is "A", i.e. which of ρ, ρ′ is called ρ) therefore matters for the **phase** of the constant c in F₁₀ (swapping it changes φ_A₅ by 0.16–0.24 rad), but never for the exponent: the exponent 1/24 + i(s₁+s₂)/10 is the same in all three fields and for either orientation.
2. **[project]** The forced central zeros of 2I live only on faithful pieces (ρ, ρ′ or Sym³ρ, which see the quaternion −1). The pieces that do not see −1 (A₅ pieces) have W = +1 and, in all nine cases, nonzero central values. So the "extra sign" of the double cover is exactly what moves the central exponent; the A₅ part only rotates c.
3. **[correction]** `icosa-oddW-aligned-exponent-20261114.md` §4 said "Since L(3′, s) is the Galois conjugate of L(3, s), they vanish together." That is not a known theorem (Galois conjugation does not act on values at s = ½ of Artin L-functions in any proved way); it should have been labelled [speculative]. It is now moot: neither vanishes in any of the three fields.

## 4. What this does and does not show

- It removes the last numerical proviso of the 2I 1/24 statement for the three fields (float level), records 6 new central values (the L(3), L(3′) pairs for 1951/2141 and galpol, plus galpol L(4)), and audits p = 2 of 2141 a second way.
- **Not** interval-certified (the Γ_ℝ(s)³ kernel has no Bessel closed form; an Arb certificate would need a Meijer-G/Mellin–Barnes enclosure, ~10⁵ terms — feasible later). Conditional on Artin holomorphy for every L-function used. Frobenius tables are PARI computations, not certified, but the 2⁻¹⁷⁵ functional-equation fits are a strong global check on them.
- No claim about zeros off the line; **no RH progress is claimed or implied.**

## 5. Files (`a5central1115/`)

`galpol_pari.py` → `galpol-pari-*-b64-n400000.json`, `-b128-n2000000.json` (float coefficients); `galpol_exact.py` → `galpol-exact-*-b160-*.json`, log `galpol-exact.log`; `quintic_check.py` → `quintic-check-b64.json`, `-b160.json`; `oddW_sym2.py` → `oddW-{three,threep}-<label>-b{64,160}-*.json` (+ `-p2flip`); `theta_indep.py` → `theta-indep-*.json`; `real_segment.py` → `real-segment.json`; `phase.py` → `phase.json`; `summary.py` → `summary.json`; `fig.py` → `../figures/a5-central-values.png/.svg`. Environment: `/workspace/work/venv-pari` (cypari2, mpmath); figure with system `/usr/bin/python3` (matplotlib). Note: cypari2 method calls (`pari.lfun(L, …)`) run at 64 bits regardless of `set_real_precision_bits`; evaluate GP strings (`pari('lfun(LG,1/2)')`) for high precision.
