# An explicit Q8 central zero: field, sign certificate, and actual L-values

[project — status update] Central simplicity for both negative examples is now interval-certified; see `q8-sprint.md` and `q8-simplicity-results.json`. Earlier statements below about simplicity being open describe the earlier numerical stage, not the current result.

[project] This continues `angles.md`; it supersedes that notebook's statement that we had not yet evaluated an explicit quaternion-group Artin L-function. It does not change the distinction between evaluating ordinary zeta at quaternion inputs and studying genuinely quaternionic arithmetic. No RH progress is claimed.

## 1. Two precisely identified arithmetic examples

[known — primary database records] LMFDB identifies these faithful two-dimensional representations of Q8. The records were opened and archived in `q8-sources/` on November 10, 2026 PT. The positive record returned a fetch-provider 403 but its same public page was readable by ordinary requests; no authentication or internal API was used.

| Record | Negative example | Positive comparison |
|---|---|---|
| Artin representation | [2.2304.8t5.b.a](https://www.lmfdb.org/ArtinRepresentation/2.2304.8t5.b.a) | [2.2304.8t5.a.a](https://www.lmfdb.org/ArtinRepresentation/2.2304.8t5.a.a) |
| Number field | 8.0.12230590464.1 | 8.8.12230590464.1 |
| Polynomial | x^8 + 12x^6 + 36x^4 + 36x^2 + 9 | x^8 - 12x^6 + 36x^4 - 36x^2 + 9 |
| Signature (real embeddings, complex pairs) | (0,4) | (8,0) |
| Discriminant | 12230590464 | 12230590464 |
| Galois group | Q8, transitive label 8T5 | Q8, transitive label 8T5 |
| Conductor | 2304 = 2^8·3^2 | 2304 = 2^8·3^2 |
| Frobenius–Schur indicator | -1 | -1 |
| Root number | -1 | +1 |
| Gamma shifts | (1,1) | (0,0) |

[project — exact arithmetic checked] PARI `polisirreducible`, `nfinit`, `galoisinit` and `galoisidentify` independently verify irreducibility, field discriminants and the full Galois groups. `galoisidentify` returns GAP SmallGroup(8,4), which is Q8; do not confuse this small-group label with the distinct transitive-group label 8T5. The character table has unique degree-two column with values 2 at the identity, -2 at the central involution and 0 elsewhere. Squaring all eight exact permutations gives Frobenius–Schur sum -8 and indicator -1. The character kernel is trivial.

[known — representation theory] This representation is symplectic/quaternionic, despite having a real-valued character. A complex representation can have real traces without being realizable by real two-by-two matrices. Its scalar Artin coefficients are real integers. Quaternionic representation type is not the same statement as quaternion-valued coefficients, or quaternion-valued inputs.

[known — Artin formalism and Hecke theory] Take a cyclic subgroup C4 of index two and its non-real faithful character λ. The degree-two character is Ind_C4^Q8 λ. Consequently its Artin L-function equals a nontrivial finite-order Hecke L-function over the quadratic fixed field. It is entire in this case; the assertion does not need the unproved general Artin holomorphy conjecture. The induction identity is Artin formalism. Primary framework references are Artin, *Über eine neue Art von L-Reihen* (1923), and Hecke, *Eine neue Art von Zetafunktionen und ihre Beziehungen zur Verteilung der Primzahlen*, first and second communications (1918/1920). Those original texts were not opened in this pass; these are framework citations, not a claim of a fresh historical-text audit.

## 2. Completion and why the sign forces a central zero

[known] Write Γ_R(s)=π^(-s/2)Γ(s/2). With N=2304 and ν=1 for the negative example or ν=0 for the positive one,

    Λν(s) = N^(s/2) Γ_R(s+ν)^2 Lν(s),
    Λν(s) = Wν Λν(1-s),                 Wν in {+1,-1}.

[known] Complex conjugation is the central involution in the negative field, so both gamma shifts are 1; it is the identity in the totally real comparison, so both shifts are 0. A negative sign forces Λ(1/2)=0 and hence L(1/2)=0, because the completing factor is finite and nonzero there. A positive sign does not force a zero. The historical quaternion-field central-zero phenomenon is established mathematics, not our discovery: Armitage, *Zeta functions with a zero at s=1/2* (1972), DOI [10.1007/BF01404125](https://doi.org/10.1007/BF01404125); Fröhlich, *Artin-root numbers and normal integral bases for quaternion fields* (1972), [primary record](https://geodesic.mathdoc.fr/item/IM_1972__17_142165/). Original full scans remain unavailable here.

## 3. A rigorous sign certificate, not just a supplied minus sign

[project — checked with ball arithmetic] `q8-certify.py` certifies W=-1 and W=+1 without inserting either root number into an L-evaluation. It independently derives every Euler coefficient through n=1000 by exact polynomial factorization modulo every unramified prime up to 1000, then encloses two theta sums with Arb. A proved explicit bound encloses the entire infinite tail. This is stronger than merely observing that software returns zero when given an odd functional equation.

[known/project — local-factor derivation] At an unramified prime, all irreducible factors of the degree-eight Galois polynomial have the same degree f. For Q8: f=1 gives trace +2, f=2 gives trace -2, f=4 gives trace 0. Determinant is 1, so the reciprocal Euler polynomial is 1-a_p T+T². The coefficients satisfy a_(p^e)=a_p a_(p^(e-1))-a_(p^(e-2)). At the ramified primes 2 and 3 the local factor is 1: inertia is nontrivial, and every nonidentity Q8 element in this faithful representation has no eigenvalue 1. Multiplicativity supplies all other a_n.

[known/project — uniform coefficient bound] The representation has finite image and can be made unitary, so its unramified Euler roots have modulus 1. The degree-two local recurrence gives |a_(p^e)|≤e+1; ramified coefficients here vanish. Therefore |a_n|≤d(n)≤n. No conjectural cancellation estimate is used in the tail bound.

[project — Mellin derivation] Define c=2πt/48 and

    Θν(t) = 4(t/48)^ν ∑_{n≥1} a_n n^ν K0(cn).

Its Mellin transform is precisely Λν(s). This follows from

    ∫_0^∞ K0(ax) x^(s-1) dx = 2^(s-2) a^(-s) Γ(s/2)^2.

[known/project] The entire self-dual functional equation implies Θν(1/t)=Wν t Θν(t), with no polar correction. Thus, for positive t, opposite signs at t and 1/t certify W=-1; equal nonzero signs certify W=+1. This use of the known analytic functional equation does not presuppose which sign it has.

[project — proved elementary tail] From the integral for K0 and cosh u≥1+u²/2,

    0 < K0(x) ≤ e^(-x) sqrt(π/(2x))        (x>0).

With r=e^(-c), m=M+1 and M=1000, set

    S2 = r^m [m²/(1-r) + 2mr/(1-r)² + r(1+r)/(1-r)³].

This equals ∑_(n>M) n²r^n. Substituting |a_n|≤n gives the valid, deliberately loose bounds

    |tail Θ1(t)| ≤ sqrt(t/12) S2,
    |tail Θ0(t)| ≤ 4 sqrt(12/t) S2.

[project — certified interval outcomes] At 256-bit Arb working precision, the resulting enclosures exclude zero by a large margin. Read the JSON for the full radii; the following coarse decimal intervals contain those enclosures:

| Example | Θ(3/4) | Θ(4/3) | Certified W |
|---|---|---|---|
| Negative | (-0.177599, -0.177597) | (0.133198, 0.133199) | -1 |
| Positive | (10.120627, 10.120630) | (7.590470, 7.590473) | +1 |

[project — explicit tail bounds] The negative example's omitted tails are below 5.715×10^-37 at t=3/4 and 2.814×10^-70 at t=4/3. For the positive example they are below 3.658×10^-35 and 1.013×10^-68. These are tail bounds, not claims that every Bessel enclosure attains that many digits. Some finite-sum ball radii are wider; all still give unambiguous signs. `q8-certificate.json` records both.

[project — scope of certification] The certificate uses exact modular arithmetic plus Arb's rigorous Bessel and elementary-function enclosures and the displayed analytic tail proof. It is reproducible computational mathematics, not a machine-checked formal proof or an algebraic product of local epsilon factors. It certifies the root-number signs, and the known functional equation then proves the exact central zero. The derivative and plotted L-values below are ordinary high-precision computations, not rigorous intervals.

## 4. Actual L-values and consistency checks

[project — computed] `q8.py` uses cypari2 2.2.4, whose wheel supplies PARI 2.17.2. It constructs each L-function from its exact field and character through `lfunartin`, with the initial root-number slot equal to 0 (unknown). No LMFDB sign is supplied. At 40, 70 and 100 decimal digits, `lfunrootres` recovers -1 and +1. Unknown-sign theta ratios at three real/complex points agree with those signs to the requested precision. Giving the functional-equation checker the wrong sign makes it fail, as it should.

| Quantity | Negative example | Positive comparison |
|---|---|---|
| L(1/2), 100-digit run | `0.E-115` | 0.708769134170846566245819950006167627268607149… |
| L'(1/2) | 3.402903132809379191539935277307563857290332709… | 1.063850001949061228190062288652019364123946825… |
| Functional-equation check, log2 error estimate | -338 bits | -343 bits |
| Largest of three unknown-sign ratio discrepancies | 1.151×10^-101 | 4.581×10^-104 |

[project — caveat] `0.E-115` is PARI's numerical-zero notation, not a separately certified 10^-115 error bound. The central evaluation uses the functional equation and is not independent evidence of a forced zero. Independent information comes from the interval sign certificate, exact field/group/local factors, unknown-sign theta ratios, negative-control checks, stable nonzero derivative and nearby noncentral values. The derivative strongly supports a simple zero numerically, but simplicity has not been certified by an interval derivative calculation.

[project — computed] On the real axis the negative example gives

    L(0.499) = -0.00340849159056708841844836992285…
    L(0.501) =  0.00339732590817275579767129182028…

These need not be exact negatives: it is the *completed* Λ, not raw L, that has odd reflection symmetry.

[project — actual plot] `figures/q8-computed-central-zero.svg` uses 201 computed samples per curve for -0.5≤t≤0.5. Let Qν(s)=N^(s/2)Γ_R(s+ν)^2. The plotted real Hardy-style normalization is

    Z_minus(t)= Im Λ_minus(1/2+it) / |Q1(1/2+it)|,
    Z_plus(t) = Re Λ_plus(1/2+it)  / |Q0(1/2+it)|.

Both are real with |Z(t)|=|L(1/2+it)|. The negative curve is odd and crosses at zero; the positive curve is even and equals 0.708769… at the centre. At t=0.1 their values are 0.339248435415184 and 0.732044706487890. The curves compare different arithmetic L-functions; their common conductor does not make their gamma factors equal.

## 5. The quaternion and octonion output distinction is now checked on real Artin data

[project — exact deduction plus computed values] These Artin coefficients are real, so their intrinsic H/O lifts obey the same slice formula. With a plus-sign mirror, a critical input sphere has real completed output. With a minus-sign mirror, it has purely imaginary completed output. The latter need not collapse to a single point. At imaginary radius 0.1, the actual computed examples give

    Λ_minus(1/2 + 0.1 I) = 0.629833709433574102210975171970296… I,
    Λ_plus (1/2 + 0.1 I) = 36.0589223011867777737397122218827… .

[project] Thus the negative example sends that entire critical S² (or S⁶) to an imaginary output sphere of radius 0.6298337094…, while the positive example sends it to one real value. At j and k, the negative outputs are genuinely different numbers, with the same norm. At the real centre, all directions have shrunk away and the forced value is exactly zero. This replaces the earlier cosh/sinh-only illustration with actual arithmetic L-function data.

[known/project — limit] This is not a change to ordinary ζ(1/2), which remains nonzero, and not an RH argument. The arithmetic root number changes which outputs are allowed; ordinary intrinsic zeta still retains its complex information. Quaternionic type alone is insufficient: both examples have indicator -1 but opposite root numbers.

[project — finite extra observation] The first 2000 exact coefficients satisfy a_minus(n)=Kronecker(-4,n) a_plus(n). This matches the expected quadratic-twist relation; the two polynomials also satisfy P_minus(x)=P_plus(ix). The finite coefficient test by itself is not a proof of a global twist identity and is not used in the sign certificate.

## 6. Reproduce and continue

[project — method that worked] The old apt attempt failed; isolated pip wheels worked without building PARI or needing system make/GMP headers. This public install route is now documented so it need not be researched again:

```sh
python3 -m venv /workspace/tools/pari-venv
/workspace/tools/pari-venv/bin/pip install cypari2==2.2.4 python-flint==0.9.0
/workspace/tools/pari-venv/bin/python q8.py
/workspace/tools/pari-venv/bin/python q8-certify.py
python3 q8-figures.py
```

[project — precision lesson] Direct cypari2 function methods may default to 53 bits unless an explicit precision is passed. An initial theta-ratio test correctly failed at ~10^-15 even though the GP global precision had been raised. The script now evaluates precision-sensitive functions through the GP evaluator at the requested precision. The test was not weakened. The independent Arb certificate does not depend on that GP floating-point precision.

[project — sources and access] The explicit LMFDB representation pages are primary computational data; PARI's installed `lfunartin`, `lfunrootres`, `lfuntheta`, `lfuncheckfeq` and `lfunlambda` documentation was read directly. An attempted official online L-function documentation path returned 404. Arb's rigorous ball arithmetic is described by F. Johansson, *Arb: Efficient Arbitrary-Precision Midpoint-Radius Interval Arithmetic* (2017), DOI [10.1109/TC.2017.2690633](https://doi.org/10.1109/TC.2017.2690633). The implemented K0 call and interval constructor were checked against installed python-flint documentation. No literature-worker files were changed.

[speculative — worthwhile next tests] Certify a nonzero central derivative with Arb to prove simplicity; derive an independent algebraic/local-epsilon sign computation; extend the certified examples to further quadratic twists; and examine critical-output sphere Jacobians and norm-level topology for the actual negative-sign function. These are precise next experiments, not proposed RH breakthroughs.


## 7. Literature-led follow-through: the explicit Magma/Armitage field

[known — primary computational sources] The [Magma handbook example H139E4](https://magma.maths.usyd.edu.au/magma/handbook/text/1641) documents the totally real quaternion field

    P(x)=x^8-820x^6+223040x^4-24206400x^2+871430400.

It gives Artin-factor conductors 1, 5, 41, 205 and 42025, with respective multiplicities 1,1,1,1,2; the faithful degree-two factor has sign -1. The same page also explicitly lists our smaller conductor-2304 negative example. The relevant LMFDB record is [2.42025.8t5.a.a](https://www.lmfdb.org/ArtinRepresentation/2.42025.8t5.a.a), field 8.8.74220378765625.1.

[project — exact arithmetic checked] `q8-magma.py` independently checks the Magma polynomial's irreducibility and reduces it to **exactly** the polynomial displayed by LMFDB:

    x^8-3x^7-63x^6+90x^5+1311x^4-20x^3-7702x^2-5524x-1009.

Its full Galois group is Q8, signature (8,0), field discriminant 205^6=74220378765625, faithful Artin conductor 42025, gamma shifts (0,0). Thus this is an additional negative-sign example with the *same gamma shifts* as our positive-sign comparison, not merely an artifact of comparing an imaginary field with a real one.

[project — independently certified sign] The script derives all coefficients through n=3000 exactly and encloses the same K0 theta sums with Arb, now using sqrt(N)=205 and ν=0. At polynomial-generator index primes (3 and 61 in this range), it uses maximal-order prime decomposition rather than incorrectly reading repeated polynomial factors as unramified Frobenius data. The independent coefficients exactly match PARI's first 3000 Artin coefficients. The general tail bound is

    |tail| ≤ 4 sqrt(π/(2c)) S2,  c=2πt/205,

with the same geometric S2 as above but M=3000. The enclosed theta values are about -1.961526900039623595 at t=3/4 and +1.471145175029717696 at t=4/3. Their tails are below 1.480×10^-20 and 3.103×10^-44, respectively. Both enclosures exclude zero, proving W=-1 by the displayed theta functional equation.

[project — computed] At 40, 70 and 100 decimal digits, the central L-evaluation is numerical zero. At 100 digits it is `0.E-117`; L'(1/2)=1.65218679617368963979579420601650174908921274334641596960137…; the functional-equation check gives -338 bits. This again supports a simple Artin zero numerically, not an interval-certified multiplicity. All roots signs start unknown; the wrong-sign control fails as expected. There are 475 checks in this additional script. The exact central zero follows from the certified negative sign and the known analytic functional equation, not from the printed numerical-zero notation.

[known/project — radical caution] Magma constructs the field using (5+sqrt(5))(41+sqrt(205)) over Q(sqrt(5),sqrt(205)). The second radical is sqrt(205), **not** sqrt(41). This is not the same literal radical as the secondary-source example rejected in `angles.md`; the newly verified polynomial and explicit normality calculation avoid relying on that questionable printed formula.

## 8. A different quaternion zeta: Hamilton/Hurwitz does not vanish at half

[known — primary author text] Voight, [*Quaternion Algebras*](https://jvoight.github.io/quat-book.pdf), Theorem 26.4.9 and §26.8 (especially Proposition 26.8.16 and Theorem 26.8.19), treats a different object: the ideal-index zeta of a maximal quaternion order. The cited statements were read in the literature worker's already downloaded, read-only author text. For the Hamilton algebra with Hurwitz maximal order and finite discriminant 2,

    zeta_B(s)=(1-2^(1-2s)) zeta(2s) zeta(2s-1).

[project — exact limit, known consequence] Its central value is **not zero**:

    zeta_B(1/2) = -log(2)/2 = -0.346573590279972654708616060729… .

Put s=1/2+h. The local factor is 2h log(2)+O(h²), zeta(1+2h)=1/(2h)+O(1), and zeta(2h)→-1/2. The local zero and zeta pole cancel, leaving that finite nonzero limit. `q8-magma.py` checks convergence from both sides at h=10^-2,10^-6,10^-12 and their negatives. Substituting zero times infinity without the analytic limit would give the wrong answer.

[known/project — do not conflate] A Q8 Artin L-function, the Dedekind zeta of its degree-eight number field, ordinary zeta at a quaternion input, and the Hurwitz-order ideal zeta are four distinct objects. The forced Artin central zero does not assert that all functions called quaternionic zeta vanish at half. Voight's completed order zeta has a plus-sign functional equation and central order m-2, with m counting all ramified places, including infinity. Hamilton has m=2 and no central zero. Those are different arithmetic constraints from the Artin root-number phenomenon.

[project — files] The extra exact identification, interval sign proof and Hurwitz-limit calculations are in `q8-magma.py` and `q8-magma-results.json`. The literature worker's files were read only; none were modified.
