# 2O aligned product: the 5/24 central exponent

Routine sprint, 13 November 2026, ~11:40 AM–12:15 PM PST. Follows `octa-central-certificate-20261113.md` (interval certificates L′(ρ, ½) ≠ 0 and L(ρ⊗ε, ½) ≠ 0) and the 2T identity in `tetrahedral-frobenius-20261112.md` / `tetra1112/certify_exponents.py`. Scripts and outputs: `octa-exp1113/`.

Labels: [known] established, with source; [project] computed here; [speculative] unproved. **No RH claim.** This writes the aligned Model A product of one 2O field as a product of Artin L-functions times a holomorphic factor, and reads off the branch of that product 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. The object

**[project]** Let K/ℚ be the galpol real Galois field of degree 48, Gal(K/ℚ) = 2O = GAP [48,28], ramified at 2, 5 and 17, as in `polyhedral-fields-20261112.md`. The group 2O ⊂ SU(2) has eight conjugacy classes:

| class | size | order | χ_ρ = 2 Re | ε |
|---|---:|---:|---:|---:|
| C1 | 1 | 1 | 2 | +1 |
| C−1 | 1 | 2 | −2 | +1 |
| C4a | 6 | 4 | 0 | +1 |
| C4b | 12 | 4 | 0 | −1 |
| C6 | 8 | 6 | 1 | +1 |
| C3 | 8 | 3 | −1 | +1 |
| C8+ | 6 | 8 | +√2 | −1 |
| C8− | 6 | 8 | −√2 | −1 |

The irreducible characters are 1, ε (sign of S₄ = 2O/{±1}), χ₂ (the 2-dimensional representation through S₄ → S₃), ψ₃ and ψ₃⊗ε (the two 3-dimensional representations of S₄), the two faithful 2-dimensional characters ρ and ρ⊗ε (Galois conjugates over ℚ(√2)), and the faithful 4-dimensional σ₄ ≅ ρ ⊗ χ₂. The table and the eigenvalues of every representation on every class are recorded and orthogonality-checked in `octa-exp1113/derive_identity.py`.

The **aligned product along an 8-fold axis**, the 2O analogue of the 2T product F₆, is the complex Euler product

    F₈(s) = ∏_{Frob_p ∈ {C1, C−1, C4a, C8+, C8−}} (1 − ψ_p p^{−s})^{−1},

with the Model A convention that each class uses a single representative in the cyclic group ⟨e^{iπ/4}⟩ ⊂ ℂ, chosen with nonnegative imaginary part: ψ = 1, −1, i, e^{iπ/4}, e^{i3π/4} respectively. The three classes that cannot be put in this slice (C4b, C6, C3) are omitted from F₈; they are absorbed into the correction H below. Ramified primes 2, 5 and 17 are removed throughout.

## 2. The exact identity

**[project, proved — computer-assisted].** For every unramified conjugacy class, the local identity

    F₈,p = L_p(ρ,s)^{1/3} L_p(ρ⊗ε,s)^{−1/6} L_p(σ₄,s)^{1/6}
           · ζ_p(s)^{i(1+√2)/8} L_p(ε,s)^{i(1−√2)/8} L_p(χ₂,s)^{i/4}
             L_p(ψ₃,s)^{i(−1−√2)/8} L_p(ψ₃⊗ε,s)^{i(−1+√2)/8}
           · ζ_p(2s)^{1/8 − i/8} L_p(ε,2s)^{−1/8 + i/8}
             L_p(ψ₃,2s)^{1/8 + i/8} L_p(ψ₃⊗ε,2s)^{−1/8 − i/8}
             L_p(ρ,2s)^{i√2/8} L_p(ρ⊗ε,2s)^{−i√2/8}
           · H_p

holds exactly, as an identity of formal Euler factors (equivalently, of power-sum coefficients for every k ≥ 1). The k = 1 layer uniquely determines the exponents of L(χ, s): the class function “ψ on the slice, 0 off the slice” expands in the character table. The k = 2 layer uniquely determines the exponents of L(χ, 2s). The remainder, starting at k = 3, is the local factor H_p, with a closed form on each class (H(0) = 1):

- C1 and C−1: H = 1
- C4a: d log H / dx = −2i x² / (1 − x⁴)
- C4b: H = (1 − x⁴)^{1/4}
- C6: H = (1 + x³)^{1/3}
- C3: H = (1 − x³)^{1/3}
- C8±: d log H / dx is an explicit rational function of degree (10, 12), whose denominator factors over ℚ(√2) as a product of (x±1), (x² ± √2 x + 1) and (x⁴ ± √2 x² + 1). Every root lies on the unit circle.

Hence H(s) = ∏_p H_p(p^{−s}) is an Euler product whose logarithm starts at p^{−3s}, and is holomorphic and zero-free for Re s > 0 (each local factor is holomorphic and nonzero for |p^{−s}| < 1). In particular H(½) ∈ ℂ^{×}.

Verified: the logarithmic-derivative identities hold as formal series through degree 48; the local Euler factors match to < 10⁻¹⁵ at x = 0.1 on every class; the global products of both sides, over the 4 200 unramified primes p ≤ 40 000, agree to relative error 2.5·10⁻¹³ at σ = 2, 8.0·10⁻¹³ at σ = 3/2, and 8.4·10⁻¹³ at σ = 1.2 (`check_product.py`).

The imaginary exponents are phases. On the real axis they contribute a constant factor exp(−Im(α) Arg L) and do not change the vanishing order of |F₈|. Compare 2T, where the analogous phase was the real-exponent ratio [L(ρ⊗ω)/L(ρ⊗ω̄)]^{1/2}. For 2O both faithful 2-dimensional characters are real-valued, so the phase is carried by imaginary powers of the S₄ L-functions.

## 3. The central exponent

**[project, proved, with one numerical hypothesis recorded below].** Along the real axis, as s → ½⁺,

- L(ρ, s) has a simple zero at s = ½, contributing (s − ½)^{1/3}. This uses the interval certificate of `octa-central-certificate-20261113.md`: the root number of ρ is −1, so Λ(½) = 0, and L′(ρ, ½) ∈ 1.94965615008486 ± 4.6·10⁻¹⁵ ≠ 0.
- L(ρ⊗ε, ½) ≠ 0 (same certificate: 4.734975816857929 ± 5.1·10⁻¹⁶), so the −1/6 power is a nonzero constant.
- ζ(2s) has a simple pole at s = ½, contributing (s − ½)^{−1/8}.
- L(ε, 1), L(ψ₃, 1), L(ψ₃⊗ε, 1) appear with real 2s-exponents ±1/8. These L-functions are automorphic (ε is Dirichlet; ψ₃ and ψ₃⊗ε factor through S₄, and ψ₃ = Ad ρ is automorphic on GL(3) by Gelbart–Jacquet 1978, Ann. Sci. ENS 11, once ρ is, which it is by Tunnell 1981). Non-vanishing on Re s = 1 is [known] (Jacquet–Shalika 1976, Invent. Math. 38). Independently, PARI gives L(ε, 1) ≈ 0.430, L(ψ₃, 1) ≈ 0.878, L(ψ₃⊗ε, 1) ≈ 1.732, all nonzero. The remaining 2s-exponents are purely imaginary.
- ζ(½), L(ε, ½), L(χ₂, ½), L(ψ₃, ½), L(ψ₃⊗ε, ½) appear only with imaginary exponents. PARI values: −1.460, 0.232, 1.020, 0.651, 1.874, all nonzero. They contribute a constant (and a choice of argument) to |F₈|.
- H(½) ≠ 0, ∞, as above.

The only remaining input is L(σ₄, ½), which appears to the power 1/6.

**[project, float numerics, not an interval certificate].** σ₄ = ρ ⊗ χ₂ has conductor 2¹⁴·5⁴·17² = 2 959 360 000, by the conductor-discriminant formula (the other seven Artin conductors are known from PARI `lfunparams`, and |d_K| = 2¹³⁶·5³⁶·17³²). The Γ-factor is Γ_ℝ(s)⁴ (K is totally real). The Dirichlet coefficients at unramified n are a_n(ρ) a_n(χ₂). Comparing the truncated Dirichlet series for L(σ₄, 2) (p-adic tail ~ 10⁻⁶) with PARI `lfun` under both signs of the root number, W = +1 matches to 4·10⁻⁷ and W = −1 is off by 6·10⁻⁵. Under W = +1, L(σ₄, ½) ≈ 1.20096 and L(σ₄, 1) ≈ 1.0398, both nonzero. (A 200 000-term evaluation against a conservative PARI cost of ~ 2.9 million terms; the value 1.201 is stable from ε = 10⁻¹ down to 10⁻⁵ in the nearby table.) This is a floating-point check, not an Arb enclosure, and the ramified Euler factor at 5 may not be exactly the product of the factors of ρ and χ₂ (χ₂ has a 1-dimensional inertia invariant at 5; ρ has none). A wrong factor at a single prime cannot create or cancel a zero of the global L-function, but it can shift the value. We treat L(σ₄, ½) ≠ 0 as a numerical hypothesis, not a theorem.

**Therefore F₈(s) ∼ c (s − ½)^{5/24}** as s → ½⁺ along the real axis, with c ≠ 0, ∞, once L(σ₄, ½) ≠ 0 is granted. The exponent is 1/3 − 1/8 = 5/24. The monodromy of F₈ once around s = ½ is 5π/12, up to the phases from the imaginary exponents and from Arg L(σ₄).

This is the 2O analogue of the 2T formula F₆(s) ∼ c (s − ½)^{3/8}. The 2T 3/8 came from L(ρ)^{1/2} and ζ(2s)^{1/8}; here the slice is a C₈ rather than a C₆, the weight on L(ρ) is 1/3 rather than 1/2, and a 4-dimensional piece appears, but it does not vanish at ½, so the net exponent is smaller: 5/24 ≈ 0.2083 against 3/8 = 0.375.

## 4. Numerical slope

**[project, float numerics].** Evaluating the closed form (PARI `lfun` of all eight L-functions, including L(σ₄) under W = +1, and the exact 2s factors) at s = ½ + ε, the consecutive slopes of log |F₈| against log ε are

| interval | slope | residual vs 5/24 |
|---|---:|---:|
| 10⁻⁵ → 5·10⁻⁵ | 0.208329 | −0.000004 |
| 5·10⁻⁵ → 10⁻⁴ | 0.208321 | −0.000012 |
| 10⁻⁴ → 5·10⁻⁴ | 0.208292 | −0.000042 |
| 10⁻³ → 5·10⁻³ | 0.207911 | −0.000423 |
| 10⁻² → 10⁻¹ | 0.201036 | −0.007297 |

The last-interval slope is 0.208329 against 5/24 = 0.208333. (H is omitted: it is holomorphic and slowly varying at ½, and does not affect the limiting slope.) Script: `octa-exp1113/slope-results.json`.

## 5. What is proved, what is numerical, what is open

- [project, proved — computer-assisted] The local Euler-factor identity for F₈, and holomorphy of H on Re s > 0.
- [project, computer-assisted, interval arithmetic] The simple zero of L(ρ) at ½ and L(ρ⊗ε, ½) ≠ 0, from the 13 Nov morning certificate.
- [known] Non-vanishing of the automorphic L-functions ε, ψ₃, ψ₃⊗ε on Re s = 1 (Jacquet–Shalika; automorphy as cited in §3).
- [project, float numerics] L(σ₄, ½) ≈ 1.201 ≠ 0, and W(σ₄) = +1. Not an interval certificate. The 5/24 exponent is a theorem of the identity plus the certificates **conditional on this non-vanishing**.
- [open] An Arb/FLINT enclosure of L(σ₄, ½), and a rigorous determination of W(σ₄) (for instance from local epsilon factors, or from a theta identity with explicit tail bounds). The conductor 3·10⁹ makes a direct copy of the 2O Mellin certificate heavier than the morning run.
- [open] The same bookkeeping for a 2I field (character field ℚ(√5), `galoisinit` unavailable). Not started in this sprint beyond recording the nine conjugacy classes.
- Not claimed: anything about zeros off s = ½, any RH or GRH statement, or a 2I central exponent.

## 6. Files

- `octa-exp1113/derive_identity.py`, `derive_identity2.py`, `identify_H.py`, `identify_H2.py`: character table, exponent solving, closed forms of H
- `octa-exp1113/check_product.py`, `check-product-results.json`, `pclass.json`: local and global Euler-product checks
- `octa-exp1113/eval_lvalues2.py`, `eval-lvalues-results.json`: PARI L-values at ½ and 1, and L(σ₄) under both root numbers
- `octa-exp1113/sig4_rootno.py`, `sig4-rootno.json`: W(σ₄) from the L(2) comparison
- `octa-exp1113/slope-results.json`: slopes of log |F₈|
- `octa-exp1113/site_figure.py`; figures `octahedral-exponent-site.svg` and `-mobile.svg`
- Python: `/workspace/tools/pari-venv/bin/python` (cypari2, sympy).

**Erratum (13 Nov 2026, ~5:05 PM PST, from `icosa-exponent-20261113.md` §2).** "H is holomorphic and zero-free for Re s > 0" in §2 overstates what was shown: each local H_p is nonzero on |p^{−s}| < 1, but the product over p converges absolutely only for Re s > 1/3, which is what is proved (and suffices at s = ½). For 2I the class mean of r₃ is nonzero, so H there has a branch point at s = 1/3; for 2O this was not checked.


## Addendum, 13 Nov 2026, ~8:55 PM PST — interval certificate for L(σ₄, ½)

The floating-point value L(σ₄, ½) ≈ 1.201 and the product formula \(a_n=a_n(ρ)a_n(χ_2)\) used in the section above are **withdrawn**. The Euler factor at 17 is \(1/(1+17^{-2s})\), not the product of the factors of ρ and χ₂. With the correct coefficients, an Arb/FLINT Mellin certificate (`cert1113/certify_sig4.py`, 128-bit) gives

- \(W(σ_4)=+1\), because \(S=\sum a_n I_{1/2}(z_n)\) lies in \(6100\pm 91\) and so excludes 0 (a self-dual real L-function has \(Λ(1/2)=(1+W)S\));
- \(L(σ_4,1/2)\) in the interval **[0.9368394678458571, 0.9510108707472682]**, which excludes 0.

PARI's float with the same coefficients is 0.9432358930, inside the enclosure. The 5/24 exponent is now conditional only on the already-certified simple zero of L(ρ) and this non-vanishing (L(σ₄) is entire unconditionally). Full write-up: `sig4-cert-2I-search-20261113.md`. No RH claim.
