# Q8 continuation: certified simplicity, admissible twists, actual sphere geometry

[project — later status] `fresh.md` and `q8-low-zeros.py/results` now interval-certify the first positive height in each base and prove complete zero counts through absolute height 2. The second and third heights below remain numerical. This notebook preserves the earlier scan’s original assurance level.

[project] This supersedes the earlier open question about simplicity in `q8.md`. Two explicit negative-sign Artin central zeros are now certified simple. The twist and nonzero-zero computations have different assurance levels, clearly separated below. Nothing here is progress on RH or a claim of a new general theorem.

## 1. Interval certificate: each central zero has multiplicity exactly one

[known] For these entire, real-type, self-dual Artin functions, Λ(s)=−Λ(1−s). The earlier independent sign certificates prove L(1/2)=0. If the derivative is nonzero, the complex analytic zero has multiplicity exactly one. The completing factor is positive and nonzero at the centre, so L'(1/2)=Λ'(1/2)/Q(1/2).

[project — interval certified] The following outward-rounded rational intervals contain the full Arb derivative balls and exclude zero:

| Representation | Certified enclosure of L'(1/2) |
|---|---|
| 2.2304.8t5.b.a | (3.32, 3.49) |
| 2.42025.8t5.a.a | (1.62, 1.69) |

[project] The earlier decimal derivatives 3.402903132809379… and 1.652186796173689… lie inside these intervals, but those decimal approximations were **not** inputs to the proof. `q8-simplicity.py` uses exact Euler coefficients previously independently checked, Arb-enclosed K0 values, interval sums on 576 cells, and two proved infinite-tail bounds. It does not call PARI L-values or derivatives. This is a reproducible numerical proof with explicit analytic estimates, not a formal proof-assistant development.

### Derivation of the certificate

[project — exact Mellin identity] Put Q=sqrt(N), ν=0 or 1, c=2π/Q, and

    Θ(t)=4(t/Q)^ν ∑ a_n n^ν K0(cnt).

This theta kernel and its Mellin transform were derived and checked in `q8.md`. Splitting the completed Mellin integral at t=1 and using the negative functional equation gives

    Λ(s)=∫_1^∞ Θ(t)[t^s−t^(1−s)] dt/t,
    Λ'(1/2)=2∫_1^∞ Θ(t)t^(−1/2)log(t) dt.

[project — finite interval enclosure] Partition [1,512] into the nine dyadic bands [1,2], [2,4], …, [256,512], each split into 64 exact dyadic cells. K0(x)>0 decreases strictly for x>0, as its positive integral representation shows. On a cell [a,b], the hull of the two Arb endpoint K0 balls encloses K0(cnt) for every t in that cell. Interval products also enclose (t/Q)^ν, t^(−1/2) and log t. Sum the signed coefficient terms, multiply by cell length and add the resulting interval integrals. No point-sampling quadrature-error assumption is used.

[project — two rigorous tails] As before, |a_n|≤d(n)≤n and K0(x)≤sqrt(π/(2x))e^(−x). For ν∈{0,1} and t≥1,

    t^(ν−1)log t ≤ t,       n^(ν+1/2) ≤ n².

Write C=8 Q^(−ν)sqrt(π/(2c)), r=e^(−c), and let S2(r,m)=∑_(n≥m)n²r^n, with the exact geometric formula in `q8.md`. The omitted coefficient tail over the whole integral is bounded by

    E_n ≤ C(1/c+1/c²) S2(r,M+1).

The t> T tail, allowing all coefficients, is bounded by

    E_t ≤ C(T/c+1/c²) r_T(1+r_T)/(1−r_T)³,
    r_T=e^(−cT).

These follow by integrating t e^(−cnt), whose integral from T to infinity is e^(−cnT)[T/(cn)+1/(cn)²], and then enlarging to a positive geometric sum. Adding both bounds overcounts some omitted terms, which is safe. We use M=1000 for the conductor-2304 field and M=1500 for conductor 42025, with T=512 for both.

[project — positive completion] Finally divide the completed-derivative enclosure by the Arb-positive number

    Q(1/2)=sqrt(Q) π^(−1/2−ν) Γ(1/4+ν/2)^2.

The resulting complete intervals above are strictly positive. In particular, each negative-sign central zero is simple. For the conductor-42025 example the two completed-derivative tails are below 5.129×10^-8 and .155735; these are not derivative errors after division, and are deliberately loose. The saved JSON includes the finite sum, both tails, completing factor, full ball and coarse verified interval.

[project — hypercomplex consequence] The intrinsic quaternionic lift has derivative Λ'(1/2) times the 4×4 identity at the real central point. It is invertible. Thus the central quaternion zero is an isolated regular point, with local Brouwer degree +1. The same argument gives a regular isolated point in eight real dimensions for the octonionic lift. This is different from a positive-radius zero sphere, whose tube degree is 2 for a simple conjugate pair. No new information about ordinary Riemann zeta has been inferred.

## 2. Eighteen admissible quadratic twists retain their signs

[known — primary source checked] Shankar–Södergren–Templier, *Sato–Tate equidistribution of certain families of Artin L-functions*, [arXiv:1507.07031](https://arxiv.org/html/1507.07031), introduction/Theorem 1.4 and **Proposition 8.7**, prove root-number constancy in the admissible Q8 quadratic-twist families. The condition used here is: D is a fundamental quadratic discriminant and gcd(D,disc K)=1. These are ordinary character twists of an arithmetic representation, not rotations of an input quaternion.

[known] Proposition 8.7 compares local epsilon factors with the square of the quadratic-character epsilon factor. The conductor is a square, the determinant is trivial, and the global quadratic-character epsilon factor is 1; under the stated coprimality hypotheses the global sign stays fixed. Their quaternionic representation families have SO(even) or SO(odd) low-zero symmetry type according to the sign—not automatically symplectic zero statistics merely because the representation has indicator −1.

[project — finite computations] We chose the six positive fundamental discriminants D=13,17,29,37,53,61, all coprime to all three base field discriminants. `lfuntwist` starts from root number 0 (unknown). For each of 18 twists, we check the new conductor N D², completing gamma shifts, the first 300 exact coefficients a_n·Kronecker(D,n), a computed root number, and the functional equation. All signs agree with the known theorem. Functional-equation discrepancy estimates range from -174 to -184 bits at 50 decimal digits. These are high-precision numerical cross-checks, not eighteen new interval root-sign proofs.

| D | New conductor, two N=2304 bases | Their signs (negative / positive base) | New conductor, N=42025 base | Its sign |
|---|---|---|---|---|
| 13 | 389376 | -1 / +1 | 7102225 | -1 |
| 17 | 665856 | -1 / +1 | 12145225 | -1 |
| 29 | 1937664 | -1 / +1 | 35343025 | -1 |
| 37 | 3154176 | -1 / +1 | 57532225 | -1 |
| 53 | 6471936 | -1 / +1 | 118048225 | -1 |
| 61 | 8573184 | -1 / +1 | 156375025 | -1 |

[project — meaningful excluded control] The existing conductor-2304 positive and negative fields are related by the non-admissible character χ_-4. Their even defining polynomials satisfy P_minus(x)=P_plus(ix). Acting on iα multiplies the original Galois action on α by χ_-4, which is the central sign in the faithful Q8 action; hence the field action is the quadratic twist. Their certified root signs are opposite. This does **not** contradict the coprime theorem: D=-4 shares the ramified prime 2. The condition is doing mathematical work.

[project — software limit] PARI 2.17.2's `lfuntwist` returned “gammafactors … not yet implemented” for negative fundamental discriminants. We did not manufacture a lower-level L-function in response. The finite table therefore covers positive discriminants only. The general known theorem is broader; negative admissible twist computations remain open for a supported independent field/representation construction.

## 3. Actual quaternion output spheres, not symmetry toys

[project — exact geometry plus numerical coefficients] For an intrinsic real-type completion, write Λ(1/2+it)=A(t)+iB(t). The quaternion lift is exactly A(t)+I B(t), for every unit imaginary direction I in S². A negative mirror sign forces A=0; a positive mirror sign forces B=0. We sampled 25 rational unit directions at each of four radii in all three fields, checking squared norms and forbidden components at 70 digits. These checks share complex evaluations; they validate the construction, not independent arithmetic data in each direction.

| Input radius t | Negative conductor-2304 B(t), output B(t)I | Positive comparison A(t), output one real point |
|---|---|---|
| .1 | .6298337094… | 36.0589223012… |
| .7 | 2.7403443438… | 21.7465577250… |
| 1.2 | 1.4748839291… | 5.8317959556… |
| 2.5 | −.1313451594… | −.6381927843… |

[project] The output sphere need not grow as the input radius grows: it expands, shrinks, collapses at a zero radius, then can return with its directions reversed. B>0 sends I to the same direction; B<0 sends it to the opposite direction. The angular scale is B(t)/t. At t=2.5 the negative example has scale −.05253806376… and reverses every direction. This angular S² orientation is not the four-dimensional tube degree; they count different maps and must not be conflated.

[project] The classic conductor-42025 example also has imaginary output spheres, with radii about 17.2812838321 at t=.1 and 54.8357913790 at t=.7. The larger numbers include its different completion. A comparison of output sizes is not a normalization-independent invariant.

## 4. Low positive zero candidates and their zero spheres

[project — numerical, not a completeness certificate] PARI's critical-line zero search with mesh divisors 16 and 32 gives the same lists through height 10 (respectively 10, 11 and 15 positive candidates for the three bases). We refine the first three candidates at 70 decimal digits and check opposite real Hardy signs at offsets ±10^-7. The largest L-value residual among the nine refined points is below 3.79×10^-75. These are strong numerical observations, not Arb-certified zero locations, a Turing count, an exclusion of even-multiplicity missed zeros, or evidence that no off-line zeros exist.

| Base representation | First reported positive height | Second | Third |
|---|---|---|---|
| 2.2304.8t5.b.a | 1.6125769007629963… | 2.6361552143853642… | 3.8068658144698460… |
| 2.2304.8t5.a.a | 1.5045488054458989… | 2.7601397479157496… | 3.6200720785364039… |
| 2.42025.8t5.a.a | 1.2647000585091833… | 2.1649457652360950… | 3.0187430189148341… |

[known/project] Each such conjugate complex zero pair becomes the whole sphere 1/2+t S² in H, or 1/2+t S⁶ in O. The negative-sign central zeros are instead isolated points (t=0), and those two points are now rigorously known to be simple. The positive example has no central zero, as its nonzero central value was checked previously. “First” here means first in the agreeing numerical searches, not a certified exhaustive ordering.

## 5. Files, reproducibility and next gates

[project] `q8-simplicity.py` / `q8-simplicity-results.json` contain the derivative certificate. `q8-sprint.py` / `q8-sprint-results.json` contain the finite twist table, zero candidates and directional samples. `q8-sprint-figures.py` generates `figures/q8-actual-output-spheres.svg` and `figures/q8-zero-sphere-radii.svg`. All scripts use the established `/workspace/tools/pari-venv`; run the earlier q8 scripts first when reconstructing their exact coefficient inputs.

[project] The routine-facing `open-questions.md` is the current queue. Central simplicity is closed for these two base examples, not a universal simplicity theorem for Q8 twists. Outstanding: certified low-zero completeness/counts, negative-discriminant computational support, central-derivative certificates for further twists, and actual norm-level tube topology. None of the literature worker's files were changed, and no routine schedule or saved prompt was modified.
