# Two icosahedral (2I) Artin L-functions with root number −1 — 14 Nov 2026, ~7:20–7:50 AM PST

Labels: [known] established, with source; [project] computed here; [speculative] unproved. **No RH claim.** This settles the sign of two functional equations. It finds, moves or excludes no zero off the line; the central zeros found here are *forced* by symmetry.

Follows `sig4-cert-2I-search-20261113.md` §3 (which left the question open). Scripts and outputs: `cert1113/oriented_2I.py`, `cert1113/lw_2I.py`, `cert1113/ram_check.py`, `cert1113/realroots.py`, `cert1113/check_models.py`, `cert1113/flip_p2.py`; data `cert1113/ap-oriented-*.json`, `cert1113/lw-*.json`, logs `orient-*.log`, `lw-*.log`.

## 1. Fixing the 5A/5B ambiguity: oriented discriminant  [project, method]

The obstacle on 13 Nov: for a Frobenius of order 5 or 10, the degree-24 stem's cycle type gives the order (5 vs 10) but not which of the two A₅ classes of 5-cycles it lies over, and that decides √5 vs −√5 in the trace. The quintic's discriminant is a square, so (disc/p) cannot split the classes.

Fix: the quintic Q₅ (LMFDB Proj_Polynomial) has disc = D² with D ∈ ℤ. Fix D once. Even permutations of the roots fix ∏_{i<j}(r_i − r_j) and odd ones negate it, so a root ordering with ∏(r_i − r_j) = +D is a "positively oriented frame", and the class of (12345) in such a frame is well defined (class A). For p unramified with Frobenius a 5-cycle, order the roots in 𝔽_{p⁵} as r, r^p, …, r^{p⁴}; in that frame Frob = (12345). Then Frob ∈ A iff ∏(r_i − r_j) ≡ +D (mod p), and ∈ B iff ≡ −D. Trace: order 5 → ζ₅+ζ₅⁻¹ = 0.618 (A) or −1.618 (B); order 10 → the negatives. Swapping A↔B globally swaps ρ ↔ ρ′ (Galois conjugates, same root number).

Details: at primes dividing the index of Z[x]/Q₅ (2, 5, 73 for 1951; 3, 29, 79 for 2141) the frame is transported to the charpoly of an integral element g, with D_g = D·∏(g(r_i)−g(r_j))/∏(r_i−r_j) fixed from complex roots (17 models; 0 disagreements between models over 158 test primes, `check_models.py`). At index primes of the degree-24 stem the cycle type is read from `idealprimedec` in a p-maximal order. **Caveat:** at p = 2, +D ≡ −D, so the test is blind; neither 1951 nor 2141 is affected in the end (for 1951 Frob₂ has order 3/6; for 2141 both choices of a₂ were run: with the Γ_ℝ(s+1)² factor only a₂ = 0.618 fits (2⁻⁵⁴); the flipped a₂ = −1.618 fails for both signs (−2 / +1), so the functional equation itself fixes a₂). Chebotarev counts to 1.2·10⁵ match the class sizes (e.g. 1109/1111/1115/1133 vs expected 1130 for each 12-element class).

## 2. Results  [project, float numerics; PARI `lfuncheckfeq`, values in log₂ of the error]

| Rep | Q24 real roots | ρ(complex conj.) | Γ-factor | checkfeq W=+1 | checkfeq W=−1 | L′(½) (W=−1) | L(2): FE vs Euler product (p<1.2·10⁵) |
|---|---|---|---|---|---|---|---|
| 2.3806401.120.a (p = 1951) | 24 | +I | Γ_ℝ(s)² | +1 (fails) | **−55** | 14.6404198 | 1.4903849665 vs 1.4903849702 |
| 2.4583881.120.a (p = 2141) | 0 | −I | Γ_ℝ(s+1)² | 0 (fails) | **−54** | 8.3882157 | 1.0725169674 vs 1.0725169686 |

So **both have root number −1 and a forced zero at s = ½** (odd order; L′(½) ≠ 0 numerically, so simple — float, not certified). With the wrong Γ-factor each fails (2141 with Γ_ℝ(s)²: −5 / 0; 1951 with Γ_ℝ(s+1)²: −5 / +1). All conditional on these Λ being entire with the Artin functional equation: Brauer gives the functional equation unconditionally, but for 1951 (an even icosahedral rep) holomorphy is Artin's conjecture, open [known]; 2141 is "totally odd" (ρ(c) = −I, not odd in the det sense), so Khare–Wintenberger does not apply either — treat both as conditional.

## 3. Why the sign is −1: where the quaternion −1 sits  [project, proved from known local facts]

[known] For a character χ of ℚ_v^×, ε(χ)ε(χ⁻¹) = χ(−1) (Tate; e.g. Tate, "Number theoretic background", Corvallis 1979, §3). 

[project] Hence: if the decomposition group D_v ⊂ 2I is abelian, ρ|D_v = χ ⊕ χ⁻¹ and W_v(ρ) = χ(−1), which is exactly the scalar by which ρ acts on rec_v(−1) ∈ D_v. Since χ(−1) = ±1 and det ρ = 1, rec_v(−1) is central, i.e. **±1 ∈ 2I**, and

  W_v(ρ) = −1  ⇔  the local Artin symbol of −1 ∈ ℚ_v^× is the quaternion −1.

- v = ∞: rec_∞(−1) is complex conjugation. For 2141 the stem is totally complex, so c = −1 and W_∞ = −1; for 1951 it is totally real, c = +1, W_∞ = +1.
- v = p, tame, p ≡ 1 (mod 5): Frobenius commutes with the cyclic inertia I (τ^p = τ), so D is abelian. rec_p(−1) is the image of −1 ∈ 𝔽_p^× in I ≅ C_e, nontrivial iff e is even and (p−1)/e is odd. Ramification (`ram_check.py`, p-maximal orders): **1951**: quintic e = 5, stem (e,f) = (2,1)²(10,1)², so I ≅ C₁₀ ∋ −1, and (1951−1)/10 = 195 is odd ⇒ W_p = −1. **2141**: stem (1,1)⁴(5,1)⁴, so I ≅ C₅ ⇒ W_p = +1.
- Totals: 1951: (+1)(−1) = −1; 2141: (−1)(+1) = −1. Both agree with §2.

The two forced zeros come from opposite ends: one from a finite prime, one from the real place. In both cases the mechanism is the same tiny quaternionic fact — the central element −1 of the unit quaternions turning up as a local symbol.

## 4. Errata to 13 Nov (`sig4-cert-2I-search-20261113.md` §3)

- "field discriminant p¹⁶" for the 1951 stem is wrong: inertia is C₁₀, which gives exponent 24 − 4 = 20 (p¹⁶ is right for 2141, inertia C₅).
- "Control: 2141² has W = +1" is wrong: W_p = +1 there, but W_∞ = −1 because complex conjugation is −1, so W = −1. The earlier analysis assumed both reps even from the saved record's `Is_Even` field; polsturm(Q24) = 0 shows ρ(c) = −I for 2141. (I do not claim the LMFDB flag is wrong; the saved field may mean something else, e.g. det even.)
- The previous trial "theta ratios drifting" was the (d/p) convention error plus, for 2141, the wrong Γ-factor.

## 5. Open
- Arb certificates of L′(½) ≠ 0 for both (conductors 3.8·10⁶, 4.6·10⁶; Γ_ℝ² and Γ_ℝ(s+1)² kernels are elementary — K₀ and e^{−x}-type — so this is cheap compared with the Γ_ℝ⁴ case).
- p = 2 blind spot of the oriented-discriminant test: lift roots to ℤ₂-unramified mod 8.
- Aligned-product exponent for these fields (like the 1/24 for the galpol 2I field): with W(ρ) = −1, the central exponent picks up +1/3 from L(ρ)^{1/3}, provided the companion ρ′ (Galois conjugate, also W = −1, also forced zero!) and Sym³ρ are accounted for — redo the bookkeeping.
- Page: fold into /critical-line-hypercomplex after #binary-icosahedral-exponent (not done this run: the overnight foliation run was due to build/deploy at 8 AM).
