# σ₄ certificate and a 2I root-number search — 13 Nov 2026, ~8:16–9:00 PM PST

Follows `octahedral-exponent-20261113.md` and `icosa-exponent-20261113.md`. Scripts and outputs: `cert1113/`. No RH claim.

## 1. Interval certificate for L(σ₄, ½) and W(σ₄)  [project, computer-assisted, interval arithmetic]

σ₄ is the faithful 4-dimensional Artin representation of the binary-octahedral (2O = GAP [48,28]) galpol field, conductor \(N = 2^{14}·5^4·17^2 = 2\,959\,360\,000\), Γ-factor \(Γ_ℝ(s)^4\). The L-function is entire unconditionally (Rankin–Selberg of two GL(2) forms that are not twists of each other; Tunnell 1981 and Gelbart–Jacquet 1978 for the GL(2) lifts). It is self-dual with real coefficients, so the root number is \(W = \pm 1\) and \(Λ(s) = W\,Λ(1-s)\).

**Euler factor at 17.** The 12:15 PM evaluation used \(a_n = a_n(ρ)\,a_n(χ_2)\). That product is wrong when \(17 \mid n\). PARI `lfunartin` of the pointwise character \(χ_ρ χ_{χ_2}\) (built in under a second) gives Euler factors \(1\) at \(2\), \(1\) at \(5\), and \(1/(1+17^{-2s})\) at 17, so \(a_{17}=0\), \(a_{289}=-1\), \(a_{83521}=1\). Exact integer coefficients to \(M=10^6\) are in `cert1113/an-sig4-M1000000.bin`, audited against 4 200 pclass primes (0 mismatches) and the ramified factors. \(\max|a_n|=16\) on this range.

**Kernel.** The inverse Mellin of \(Γ_ℝ(s)^4\) is
\[
φ(x) = 16 \int_0^∞ K_0(v)\,K_0(4π^2 x/v)\,dv/v,
\]
and \(J_{1/2}(z) = \int_z^∞ φ(x)\,x^{-1/2}\,dx = (8/π)\int_0^∞ K_0(v)\,F(4π^2 z/v)\,v^{-1/2}\,dv\), where \(F(w)=\int_w^∞ K_0(t)\,t^{-1/2}\,dt\) is tabulated by chained `acb.integral`. Then \(I_{1/2}(z)=z^{-1/2} J_{1/2}(z)\) and \(Λ(1/2)=(1+W)\sum a_n I_{1/2}(n/\sqrt{N})\). Independent SciPy quadrature of the same formula gives \(J(1.5319·10^{-5})=32.56919\), inside the Arb ball \([32.569 \pm 4.72·10^{-4}]\). The gap from \(J(0)=Γ_ℝ(1/2)^4 \approx 55.001\) is the (large) small-\(x\) mass of \(φ\), as expected from the order-4 pole of \(Γ(s/2)^4\) at \(s=0\).

**Results** (`certify_sig4.py`, 128-bit, \(M=250\,000\), 2 000 F-points, 800 z-points, 154 s):

- \(S = \sum a_n I_{1/2}(z_n) \in 6100 \pm 91\), which **excludes 0**.
- Hence \(W\) cannot be \(-1\) (that would force \(Λ(1/2)=(1+W)S=0\)), so **\(W(σ_4)=+1\)**.
- \(L(σ_4, 1/2) = 2S\big/\bigl(N^{1/4} Γ_ℝ(1/2)^4\bigr)\) lies in
  **\([0.9368394678458571,\; 0.9510108707472682]\)**
  (radius \(7.1·10^{-3}\)). This **excludes 0**.
- A supporting theta test at \(y=1.2\) gives ratio \(1 \pm 0.311\), which also excludes \(-1\), but is much looser than the \(S \neq 0\) argument.
- PARI `lfun` with the same coefficients (warning: `#an = 10^6 < 2.9·10^6`) gives the float \(L(1/2) \approx 0.9432358930\), inside the enclosure.

**The earlier float 1.201 is withdrawn.** It used the wrong factor at 17. A truncated Mellin grid had previously produced the untrustworthy float 0.943; with the correct kernel that value is now confirmed and enclosed.

The 5/24 exponent in the 2O aligned product is unchanged in form: it requires \(L(σ_4,1/2)\neq 0\), which is now an interval certificate, and \(L(σ_4)\) is entire with no appeal to Artin's conjecture.

## 2. L′(Sym³ρ, ½) for 2I  [open]

Not certified this sprint. The same \(Γ_ℝ^4\) kernel applies (conductor \(2^{16}·11^4·23^2\), \(M=5.5·10^6\) coefficients already used for the float \(L'(4',1/2)\approx 14.9739\) in `icosa-exp1113/lambda_sym3.py`). An Arb enclosure that excludes 0 remains open. The odd order of the central zero is already unconditional (Brauer functional equation, \(W(4')=-1\)).

## 3. A 2I field with W(ρ) = −1  [project, incomplete]

Twelve saved LMFDB 24T201 records (Sign not stored). Odd-conductor fields:

| Baselabel | Conductor | Bad primes | \(p \bmod 20\) | \(χ(-1)\) for faithful order-10 characters of \(\mathbb{F}_p^\times\) |
|---|---|---|---|---|
| 2.3806401.120.a/b | \(1951^2 = 3\,806\,401\) | 1951 | 11 | **−1** (all \(j\) coprime to 10) |
| 2.4583881.120.a/b | \(2141^2 = 4\,583\,881\) | 2141 | 1 | **+1** (all such \(j\)) |

Both representations are even (`Is_Even`), so \(W_∞=+1\). The only ramified prime of the degree-24 stem (field discriminant \(p^{16}\), from `nfdisc` after `polredbest`) is that odd prime, so \(W(ρ)=W_p\).

For a tame 2-dimensional representation \(ρ=χ\oplus χ^{-1}\) of conductor \(p^2\) with \(p\equiv 1\pmod{10}\) (residue degree 1, inertia \(C_{10}\)), a standard identification of the Tate local constant gives \(W_p(ρ)=χ(-1)\). Under that identification, **the field of conductor \(1951^2\) has \(W(ρ)=-1\)** (forced central zero), and the field of conductor \(2141^2\) has \(W(ρ)=+1\).

**Numerical confirmation is not finished.** Cycle types of the degree-24 stem on 24 letters leave the two order-5 and two order-10 classes of 2I ambiguous (same cycle type). The quintic discriminant is a square (Galois group \(A_5\)), so \((d/p)\) is identically \(+1\) and does not split those classes. A \(\mathbb{Q}(\sqrt{5})\)-valued class invariant (the 3-dimensional character of \(A_5\)) is still needed before a theta test or `lfun` evaluation can be trusted. A trial that assigned all 5-cycles to a single class produced theta ratios drifting with \(y\), as expected from that error.

Scripts: `cert1113/search_2I_oddW.py`, `cert1113/ramification_2I.py`. The degree-60 splitting fields of the two 1951-quintics are in `cert1113/splitting-2.3806401.120.*.txt`.

## Files

- `cert1113/certify_sig4.py`, `certify-sig4-results-p128-M250000.json`, `an-sig4-M1000000.bin`
- `cert1113/search_2I_oddW.py`, `search-2I-oddW-results.json`
- This note; updates in `open-questions.md` and an addendum to `octahedral-exponent-20261113.md`.

**Erratum (14 Nov 2026, ~7:50 AM PST):** §3 is superseded by `icosa-oddW-20261114.md`. Both fields have W = −1 numerically; the 1951 stem has inertia C₁₀ (disc exponent 20, not 16); the 2141 "+1 control" ignored that complex conjugation is −1 there (W_∞ = −1).
