# Q8 aligned Frobenius product = a "half" Artin L-function (with branch points)

Sprint run, 11 November 2026, ~8:20–9:00 PM PST. Closes the [speculative] line in open-questions item 5 remainder
("the aligned-class product looks like half of a Hecke L-function"). Follows `frobenius-complex-s-20261111.md` (Model A).
Labels: [known] established mathematics; [project] proved or checked here (elementary, not deep); [speculative] unproved.
Nothing here is RH progress: every zero involved is a zero of an ordinary, classical L-function.

## Setup
[known] K/Q Galois with group Q8, rho the 2-dim faithful representation. For a quaternion unit u in {i,j,k}, let F be the quadratic
subfield fixed by <u>; Gal(K/F) = <u> is cyclic of order 4, rho = Ind_F^Q psi with psi the order-4 character u -> i, so L(rho,s) = L(F,psi,s).
chi_u, chi_v, chi_w are the three quadratic characters (chi_u has kernel <u>, i.e. it belongs to F). Primes with Frob in {±v, ±w}
are inert in F ("orthogonal" primes); primes with Frob in {±1, ±u} split in F.

Model A, aligned: s = sigma + t·u (the complex unit of s is u itself), and the "all-plus" choice g_p = +u for Frob = ±u.
[project] This choice is canonical once Gal(K/Q) is identified with the unit group: it means "take the prime P of F above p with
Frob_P = u in the abelian group Gal(K/F)", i.e. psi(P) = +i. Everything then lives in the slice C_u ≅ C (checked: the quaternion product
stays in C_u and equals the complex product to 1e-14). Model A's norm splits as

    nrd E_A(s) = |F(s)|² · O(sigma),   F(s) = ∏_{p split in F} (1 − psi(P_p) p^{-s})^{-1},   O(sigma) = ∏_{p inert in F} (1 + p^{-2 sigma})^{-1},

the second factor being the earlier observation that orthogonal primes lose all t-dependence.

## Results

1. [project — exact; series identity checked symbolically in Q(i) for all five classes; exponents are the unique solution of the linear system]

       F(s) = L(rho,s)^{1/2} · [ zeta(s) L(chi_u,s) / (L(chi_v,s) L(chi_w,s)) ]^{i/4} · [ zeta(2s) / L(chi_u,2s) ]^{1/4} · G(s),

   all L-functions partial (ramified Euler factors removed; rho has none, being faithful and irreducible), G an Euler product whose local
   logs start at p^{-3s}, so G is holomorphic and zero-free on Re s > 1/3. In words: one factor per split rational prime (instead of
   Artin's two) gives exactly the square root of the Artin L-function, times an imaginary power i/4 of a ratio of Dirichlet L-functions,
   times a quarter power of zeta(2s)/L(chi_u,2s). Companion identity: with the "all-minus" choice F⁻(s) = conj F(conj s) and
   [project, exact] L(rho,s) = F(s) F⁻(s) ∏_{inert}(1 + p^{-2s})^{-1}: Artin's L really is two "halves" plus the inert primes.
   The i/4 exponent is the same pure (i,j,k) part that showed up in the explicit formula ( ¼(ζ + χ_u − χ_v − χ_w), q8-explicit-formula note).

2. [project — float numerics, PARI lfun with continuous branches] Checked at sigma = 2 and 3/2, t in [0,40] step 0.05, for 2.2304.8t5.b.a,
   2.42025.8t5.a.a (root number −1) and 2.2304.8t5.a.a (+1), each with u = i, j, k: |log F_X − formula| ≤ 2.6e-7 (sigma 2) and
   ≤ 1.4e-4 (sigma 3/2) at X = 80,000, falling like the Euler-product tail (×4 per ×4 in X at sigma 2, ×2 at sigma 3/2). The naive guess
   F ≈ L(rho)^{1/2} is off by 0.016–0.40, so the extra factors are needed. See halfhecke1111/half-hecke-results.json.

3. [project — consequence of 1; checked by argument tracking] F continues to Re s > 1/3 only as a **multivalued** function. Its branch
   points are zeros/poles of the classical L-functions:
   - a simple zero of L(rho) multiplies F by −1 (no change in |F|);
   - a zero of zeta or L(chi_u) multiplies |F| by e^{−π/2} ≈ 0.208; a zero of L(chi_v) or L(chi_w) by e^{+π/2}; the pole of zeta at 1 by e^{+π/2}.
   So the continued Model A norm |F|²·O is path-dependent by factors e^{±π} ≈ 23 per enclosed zero.
   Check (2.2304.8t5.b.a, u = i, box 0.4 ≤ sigma ≤ 2, 10 ≤ t ≤ 20): windings 15 (rho), 1 (zeta), 5, 6, 5 (chi_u, chi_v, chi_w), equal to
   PARI lfunzeros counts; measured Δ log F = 5π/2 + 15πi/2 exactly as predicted (|F| changes by e^{5π/2} ≈ 2576 going round the box).

4. [project — the "forced zero at ½" changes character] Near s = 1/2 the quarter power of zeta(2s) (pole at 2s = 1) combines with the zero of
   L(rho): F(s) ~ c·(s − ½)^{+1/4} when the root number is −1 (L(rho,½) = 0 to order 1, PARI lfunorderzero), and F(s) ~ c·(s − ½)^{−1/4}
   when it is +1. Measured slopes of log|F| vs log(s − ½): 0.24997 and −0.24997 at s − ½ = 1e-5 (figure, panel b). Going once round s = ½
   multiplies F by i (checked: Δ arg = π/2 exactly). So in the aligned quaternion model the forced Artin zero is no longer a zero but a
   fourth-root branch point, and for root number +1 the product even blows up at ½.

5. [project — elementary, proved; numerically checked] The orthogonal factor O(sigma) converges for sigma > ½ and
   O(sigma) = [L(chi_u,2 sigma)/zeta(2 sigma)]^{1/2} · H(sigma), H zero-free for sigma > 1/4. Hence O(sigma) ~ C·(2 sigma − 1)^{1/2} → 0
   as sigma → ½⁺, at every height t. Checked for 2.2304.8t5.b.a, u = i: the formula matches direct products to 3e-8 at sigma = 1 (p ≤ 10^7);
   log O − ½ log(2 sigma − 1) → 0.447 as sigma → ½ (figure, panel a). The direct truncated product converges too slowly to show this near ½,
   so that limit rests on the formula. Interpretation: the critical line appears here as a **barrier of the Model A norm itself**, caused by
   the inert primes losing their phases (Σ p^{-2 sigma} diverges at ½), not by any zeros. This is the "tiny way" the extra quaternion
   directions change the outcome: they kill oscillation for half of the primes (Frob in {±v, ±w}, Chebotarev density 4/8; corrected ~9:05 PM PST from "one third", which miscounted classes; the exponent ½ in (2σ−1)^{1/2} is exactly this density) and the line ½ becomes a square-root branch line.

## Not claimed
No zeros of any classical L-function are found, moved or excluded; no RH information. [speculative, not checked] F probably has a natural
boundary at Re s = 0 or earlier (Estermann/Dahlquist/Kurokawa-type results on Euler products with non-integer exponents suggest this;
sources not re-read in this run). Floats, not intervals. Ramified-prime bookkeeping follows PARI's local factors.

## Files
`halfhecke1111/half_hecke.py` (exact series + three fields × three units; 17,684 checks passed, ~9 min), `half-hecke-results.json`, `run.log`;
`halfhecke1111/part2.py` (orthogonal factor, monodromy, central exponent), `part2-results.json`, `part2.log`;
`fig_data.py`, `fig_plot.py`, figure `figures/q8-half-hecke.png/.svg`.

## Page status
Not folded into /critical-line-hypercomplex: a live session was working in n4proof/ and dump jobs at 8:00–8:20 PM PST, so research and notes
only. Ready as a short subsection after #q8-frobenius-complex-s (closed form, results 3–5, one figure).

Update ~9:05 PM PST (Grant pt216u): folded into /critical-line-hypercomplex#q8-half-l-function (plain-language subsection after
#q8-frobenius-complex-s; site figure `figures/q8-half-hecke-site{,-mobile}.svg` from `halfhecke1111/site_figure.py`). Labels on the page:
identity and consequences [project]; natural boundary [speculative]; no bearing on RH. Correction made while writing it up: the
phase-losing primes are half of all primes (density 4/8), not a third.
