# Fresh results: low-zero certificates, infinity, octonion parentheses and E8

[project] November 10, 2026 PT. Four explicitly separated objects: Q8 Artin functions; the intrinsic xi zero-set complement; constructed octonion star polynomials; E8 lattice Epstein zeta. No RH-progress claim or claim of new general mathematics.

## 1. First positive Artin zeros: now bounded and counted

[project — interval proof] The first positive critical-line heights lie in the following rational intervals, each of width 10^-12:

| Representation | Lower bound | Upper bound | All completed zeros below absolute height 2, multiplicity counted |
|---|---|---|---|
| 2.2304.8t5.b.a | 1.612576900762 | 1.612576900763 | 3 |
| 2.2304.8t5.a.a | 1.504548805445 | 1.504548805446 | 2 |
| 2.42025.8t5.a.a | 1.264700058509 | 1.264700058510 | 3 |

[project] The negative-sign counts are one simple central zero plus one conjugate nonreal pair. The positive-sign count is one conjugate pair and no central zero. The certified sign changes put the pair on Re(s)=1/2; the independent total count leaves no room for a missed even-multiplicity zero or off-line zero in this finite height window. Each first positive zero is simple. The second and third heights from the previous scan remain numerical-only.

### Exact analytic identity and tail budget

[known/project] These particular Artin representations are induced from nontrivial characters of an order-four subgroup; their completed functions are entire by the corresponding Hecke theory, not by an assumption of the general Artin conjecture. Their previously independently certified signs are W=±1. Use the earlier exact Euler coefficients and put Q=sqrt(N), nu=0 or 1, c=2pi/Q,

    Theta(u) = 4(u/Q)^nu sum_(n>=1) a_n n^nu K0(c*n*u),
    Lambda(s) = integral_1^infinity Theta(u)[u^s + W u^(1-s)] du/u.

On the critical line the real Hardy-type component is

    W=+1: H(t)=Lambda(1/2+it)=2 integral Theta(u)u^(-1/2)cos(t log u) du,
    W=-1: H(t)=Lambda(1/2+it)/i=2 integral Theta(u)u^(-1/2)sin(t log u) du.

[project — proof of uniform tails] The contour uses -1/2<=Re(s)<=3/2, |Im(s)|<=2. The two powers in the integral have absolute sum at most 2u^(3/2). From |a_n|<=d(n)<=n and K0(x)<=sqrt(pi/(2x))exp(-x), the omitted integrand is bounded by

    C sum n^2 u exp(-c*n*u),   C=8 Q^(-nu)sqrt(pi/(2c)).

The same bound also bounds the Hardy integrals. With S2(r,m)=sum_(n>=m)n^2 r^n, M the coefficient cutoff and T=2048, safe bounds are

    E_n = C(1/c+1/c^2) S2(exp(-c),M+1),
    E_T = C(T/c+1/c^2) r_T(1+r_T)/(1-r_T)^3,
    r_T=exp(-cT).

They follow by integrating u exp(-c*n*u) exactly and then enlarging positive sums. Adding both errors overcounts their overlap and is safe. We use M=1000 for the two conductor-2304 fields and M=3000 for the classic conductor-42025 field. Complete bounds are saved in `q8-low-zeros-results.json`.

[project — validated signs] Arb/acb analytic quadrature at 256 bits integrates the finite Hardy kernel on eleven dyadic intervals [1,2],…,[1024,2048]. The callback checks Re(u)>0 and forwards the analytic flag to log(u), so its logarithm and K0 stay off their cuts. A finite output alone is not accepted as a sign: the full real ball, enlarged by BOTH tails, must exclude zero. The rational left and right endpoint values have opposite signs. No PARI L-values are inputs. Earlier numerical seeds choose intervals; validated bounds prove the sign changes.

### Independent total count: argument principle, not a fine search mesh

[project — proof] We enclose Lambda on the complete counterclockwise rectangle [-1/2,3/2] × [-2,2]. Each side is split into dyadic segments: 192 total. For every segment, use interval Riemann sums in u: 64 cells per dyadic band, 704 cells total. Monotonicity of K0 encloses each kernel between its endpoint balls; interval arithmetic encloses every complex exponential throughout the entire parameter segment. Add the uniform tails to both output coordinates. Every resulting convex complex rectangle excludes zero.

[project] Adjacent output rectangles both contain the true common endpoint value. Choose a polygon vertex in their intersection and verify it lies in both. Each true image arc and its polygon edge lie inside the same convex zero-free rectangle; linear interpolation gives a zero-avoiding homotopy of the closed curves. Therefore their windings agree. Sum the principal arguments of consecutive vertex ratios, checking each increment lies strictly between -pi and pi. The resulting interval divided by 2pi contains a unique integer: 3, 2, 3. All image boxes and polygon vertices are saved, and the outward-serialized boxes are also checked to exclude zero.

[known/project] The argument principle counts zeros with multiplicity because Lambda is entire and nonzero on the contour. Absolute Euler convergence makes Lambda nonzero for Re(s)>1; the functional equation does the same for Re(s)<0. Hence this rectangle covers ALL its zeros with |Im(s)|<2. The already certified simple central zero when W=-1, and the validated conjugate sign-change pair, exhaust the total. This is the completeness-comparison role requested by “Turing-style”: it is an independent total count matched to found zeros, NOT an implementation of Turing's averaged-argument inequality. PARI's documented naive `lfunzeros` mesh alone cannot supply this certificate.

[project — limits] The statement concerns three explicit Artin L-functions through height 2 only, not ordinary zeta or a family-wide GRH proof. Rebuilding the certificate requires the earlier exact arithmetic records. The underlying analytic estimates are written here; it is not a proof-assistant formalization. Arb reference: Fredrik Johansson, “Arb: efficient arbitrary-precision midpoint-radius interval arithmetic,” arXiv:1611.02831. PARI's explicit scan warning: https://pari.math.u-bordeaux.fr/dochtml/html-stable/_L_minusfunctions.html#lfunzeros.

## 2. Infinity: a nontrivial end system with trivial ordinary limit

[project — exact topology] See `infinity.md` for the full proof. The cofinal exterior groups are free groups on the surviving tail punctures, with injective bonding maps retaining meridians. The pro-system is not pro-trivial: every smaller infinity neighborhood still has an essential meridian. Yet its ordinary inverse limit is trivial: a fixed finite reduced word cannot belong to every successively deleted tail. There is no contradiction. Other ends approach individual finite zero spheres; we are explicitly studying the ambient-infinity end.

[project] The compactified zero set fails complement-side 1-LCC and is not locally connected at infinity. It is not an Alexander horned sphere or even a topological sphere there. In our complement, meridian holes survive in H1. In Alexander's example the exterior pi1 is nontrivial but H1 vanishes, by Alexander duality for its embedded topological S2. This is a genuine distinction in mechanisms, not just a visual resemblance.

[known] Primary Alexander paper and Hatcher Corollary 3.45 were inspected; see `infinity.md`. `infinity.py/results` supplies 22 finite inversion/winding/word checks, not numerical proof of the infinite statement.

## 3. Octonions: parentheses change star zeros, NOT finite pointwise zero sets

[known/project — exact reproduction] Ghiloni–Perotti, arXiv:0904.2667, **Remark 3**, uses f(w)=w e1-e2 and a right constant e4 outside H=<1,e1,e2,e3>. In our declared convention e3=e1e2, e5=e1e4, e6=e2e4, e7=(e1e2)e4. Then f has zero e3, while f*e4=w e5-e6 has zero -e3. The pointwise product f(w)e4 retains the old zero e3. As an associative control, restricting w and the right constant e3 to H gives the same pointwise/star product and unchanged zero.

[project — constructed toy, not an arithmetic L-function] Set A=1-w e1, B=1-w e2, C=1-w e4, and use the coefficient-convolution star product. Direct exact rational arithmetic gives

    P_left=(A*B)*C: zero w=(-e1-e2+e4+e7)/2,
    P_right=A*(B*C): zero w=-e1.

Both are pure imaginary units. The polynomials differ in the cubic coefficient because (e1e2)e4=e7 whereas e1(e2e4)=-e7. Their normal polynomials are exactly (1+w^2)^3, so their only zero-bearing conjugacy sphere is the SAME Re(w)=0, |w|=1 sphere. On it each remainder is A0+w B0 with B0 nonzero, proving exactly one isolated zero in each bracketing. The directions differ, not the real coordinate or radius. Translating w=q-1/2 puts both on Re(q)=1/2, with no off-line implication.

[known/project — a necessary correction to the proposed experiment] A finite POINTWISE product of octonions cannot change its zero set just by changing parentheses: norm multiplicativity gives |(ab)c|=|a(bc)|=|a||b||c|. Either is zero iff one factor is zero. For the three factors above that zero union is {-e1,-e2,-e4}. At real w=1, the two pointwise values differ but both have norm squared 8. Only the explicitly defined STAR polynomial has the bracket-sensitive isolated zero directions. Infinite products require their own convergence hypotheses; this finite argument does not silently cover them.

[project] `octonion-products.py/results` has 431 exact rational checks, including norm multiplicativity, alternativity, an associative control, published-example values, both toy roots and both normal polynomials. Source: https://arxiv.org/abs/0904.2667, Remark 3 and Lemma 7/Corollary 8. No arithmetic meaning is claimed for this finite Euler-like polynomial toy.

## 4. E8: the centre belongs to the dimension AND the exponent convention

[known/project] In the standard Euclidean E8 normalization, minimum squared norm is 2. We enumerated both coordinate cosets directly: integer coordinates with even sum and half-integer coordinates with even sum. Squared norms 2,4,6 have counts 240,2160,6720. Integer/half-integer contributions are respectively [[112, 1136, 3136], [128, 1024, 3584]]. These agree with 240 sigma3(m), independently of using that formula for enumeration.

[known] E8's theta series is E4=1+240 sum sigma3(m) exp(2pi i m tau). Thus

    Z_E8(s)=sum_(v!=0)|v|^(-2s)=240*2^(-s)*zeta(s)*zeta(s-3),
    Lambda_E8(s)=pi^(-s)Gamma(s)Z_E8(s)=Lambda_E8(4-s).

The defining positive series converges for Re(s)>4; central values use analytic continuation. Primary lattice definition and E4 expansion: Viazovska, arXiv:1603.04246, introduction and §3, freshly inspected. The theta identity itself is classical; author exposition inspected in the literature map: Nebe's 2009 Lattices and modular forms slides. Poisson summation for a self-dual n-dimensional lattice gives s<->n/2-s and centre n/4. For a non-self-dual lattice the functional equation pairs it with its dual; a self-equation is not automatic.

[project — numerical checks] At 80 digits, four complex reflected values agree to better than 10^-70. A separately evaluated theta-Mellin continuation agrees with the zeta-factor formula to better than 10^-70. The theta transform t^(-4)Theta(1/t)=Theta(t) is checked at three positive t. These floating-point checks illustrate the known exact identities; they are not Arb certificates. At the centre, the exact completed value is Lambda_E8(2)=-5/6, reproduced numerically: insert Gamma(2)=1, zeta(2)=pi²/6 and zeta(-1)=-1/12 into the factor formula. It is NOT a forced central zero.

[known/project — corrected dimension table] For |v|^(-2s): Gaussian Z2 has centre 1/2, the four-coordinate Lipschitz-quaternion lattice Z4 has centre 1, E8 has centre 2. These are lattice Epstein functions, not quaternion-order ideal-counting zetas. Changing the exponent to |v|^(-8u) gives Ztilde(u)=Z_E8(4u), so its completed mirror becomes u<->1-u and the centre is 1/2. This is a change of variable, not a new zero-location theorem or a metric rescaling of the lattice.

[known/project — essential warning] Even with its centre renamed 1/2, E8 does NOT acquire “all nontrivial zeros on 1/2.” Its two Riemann-zeta factors yield families from zeta(4u) and zeta(4u-3). Under RH, their real parts would be 1/8 and 7/8. Mirror centre and zero line are different assertions. The dimension change really changes the arithmetic object; renormalizing its variable does not erase that.

## Reproduce and continue

[project] Run q8-low-zeros.py and octonion-products.py with /workspace/tools/pari-venv/bin/python; run e8.py and infinity.py with system python3 (mpmath environment). `fresh-tests.py` asserts the delivered results. `fresh-figures.py` generates four SVGs and PNGs. The earlier q8 scripts generate the exact coefficient input files. Current open questions are in `open-questions.md`; the existing routine itself was not modified. Literature folder remained read-only.
