# Twisted Q8 central zeros are simple (12 admissible twists)

Run: hypercomplex-critical-line-sprint, Nov 10, 2026, 7:30 AM PT. Closes open item 3 ("apply the central-derivative certificate to selected admissible twists") for every positive admissible twist already in the table.

## Question

The two sign −1 Q8 Artin bases (2.2304.8t5.b.a and 2.42025.8t5.a.a) have a forced zero exactly at s = 1/2, and we had proved both are simple. Admissible quadratic twists keep the sign −1, so each twist also has a forced central zero. Is that zero simple, or can twisting make it a double zero?

## Facts used

- [known] Twisting by a real quadratic character keeps the representation quaternionic (Frobenius–Schur indicator of ρ⊗χ_D is that of ρ, since χ_D is real), and Shankar–Södergren–Templier, arXiv:1507.07031, Prop. 8.7: admissible twists (fundamental D coprime to the field discriminant) preserve the root number. Root −1 forces L(1/2)=0 [known, functional equation].
- [project, earlier run] Root numbers −1 for D = 13, 17, 29, 37, 53, 61 on both bases (q8-sprint-results.json).

## Method [project]

Λ_D'(1/2) = 2 ∫_1^∞ θ_D(x) x^(−1/2) log x dx with θ_D(x) = 4 (x/Q_D)^ν Σ χ_D(n) a_n n^ν K0(2πnx/Q_D), Q_D = D·Q (Q = 48 or 205; ν = 1 or 0). Arb (256 bits), Gauss–Legendre order 48 on dyadic bands with the analytic-circle error bound of q8-extended.py (the |sin(t log u)| factor replaced by a bound on |log u|), explicit coefficient tail and integration tail as in q8-simplicity.py. Cutoffs scale with D: up to 139,316 coefficients and integration to 262,144. Script: twist-simplicity.py; data: twist-simplicity-results.json; log: twist-simplicity-run.log; figure: figures/q8-twist-slopes.svg.

Coefficient provenance: base a_n from PARI 2.17.2 `lfunan` on the Artin representation; they agree exactly with the earlier independently audited prefixes (2000 for the 2304 base, 5000 for the 42025 base). Beyond those prefixes they are PARI's exact integer coefficients, not separately audited. Twisted coefficients are exact: a_n·kronecker(D,n).

Validation: the same script reproduces the two base slopes already certified (3.4029…, 1.6521…) to 16+ digits, and every twist enclosure contains PARI's floating value (twist-derivs-pari-results.json, which also lists L''(1/2) for reference).

## Result [project · interval proof, conditional on PARI coefficients beyond the audited prefix]

| base | D | L′(1/2) enclosure (rounded) |
|---|---|---|
| 2.2304.8t5.b.a | 1 | 3.402903132809379 ± 5e-17 |
| 2.2304.8t5.b.a | 13 | 6.825235272738845 ± 1e-16 |
| 2.2304.8t5.b.a | 17 | 10.37253806550589 ± 9e-17 |
| 2.2304.8t5.b.a | 29 | 6.898505766653101 ± 2e-16 |
| 2.2304.8t5.b.a | 37 | 11.06579146052266 ± 2e-16 |
| 2.2304.8t5.b.a | 53 | 13.76772981082526 ± 3e-16 |
| 2.2304.8t5.b.a | 61 | 13.01389179989316 ± 5e-16 |
| 2.42025.8t5.a.a | 1 | 1.652186796173690 ± 2e-19 |
| 2.42025.8t5.a.a | 13 | 4.258527531603546 ± 2e-14 |
| 2.42025.8t5.a.a | 17 | 3.998454871915128 ± 6e-14 |
| 2.42025.8t5.a.a | 29 | 4.144082579408579 ± 9e-13 |
| 2.42025.8t5.a.a | 37 | 4.997128969611346 ± 3e-12 |
| 2.42025.8t5.a.a | 53 | 4.346361829918009 ± 2e-11 |
| 2.42025.8t5.a.a | 61 | 5.940103536306276 ± 4e-11 |

All 14 enclosures exclude 0, so all 12 twisted forced central zeros are simple (order exactly 1). Total compute ≈ 13 minutes.

## Reading

- [project] In the quaternion lift, each of these functions has a single real point zero at 1/2 (degree-1), not a zero sphere; the twist changes the slope there (1.65 to 13.8) but never flattens it to a double zero in this finite table.
- [speculative] Slopes on the 2304 base grow roughly with D and those on the 42025 base stay near 4–6; this is 6 data points per base, not a trend claim. The usual "minimalist" expectation for families of L-functions is that most root-number −1 members vanish to order exactly 1 at the centre, consistent with what we see, but we do not test that here.
- Not RH progress, not a statement about all quaternionic Artin L-functions, and not about negative D (still blocked; see open-questions item 2).
