# The critical line in quaternion and octonion space

Research notebook · 9 November 2026 PT (continued across midnight if needed).

**[known]** A statement established in the cited literature, or an elementary consequence proved here. **[project]** A new-to-this-project derivation, visualization or finite experiment, not a novelty claim. **[speculative]** An unproved proposed research direction. These labels are not claims of RH progress. Every substantive paragraph below has a label; formulas inherit their paragraph's label.

## 0. What this exploration actually found

**[project]** The useful outcome is a dictionary plus several filters that stop misleading analogies. We can formalize the intrinsic slice lift and its zero-sphere degree. We cannot obtain new zeta information merely by adding imaginary directions. The genuinely different algebra appears when octonionic coefficients use three independent directions; a published example below shows exactly what changes, and exactly what does not.

**[project]** Corrections to tonight's earlier notes: (1) the simple mirror law and real values on the critical hyperplane concern completed **xi**, not zeta; (2) RH does NOT assert simple zeros, and raw sign changes cannot count even-multiplicity zeros; (3) degree around a ball counts both conjugate complex zeros, so it is twice the upper-half-plane count, not twice the count including both halves; (4) the normal function of a non-real-coefficient slice function has real coefficients, not necessarily real *values* off the real axis; (5) Artin's theorem does not make three arbitrary octonions associative; (6) the special value 1/2 comes from the functional equation, not from centrality—all real numbers are central and fixed by algebra automorphisms; (7) the fact that these zero sets are tame follows from their explicit sphere formula, not the false general claim that smooth functions cannot have wild zero sets.

## 1. The precise construction: no hidden eight-dimensional zeta

**[known; GP11, GP09]** Let A be C, H or O, of real dimension d=2,4,8. Write q=x+yI, with y=|Im q|>=0 and I a unit pure imaginary. Then I²=-1 and I belongs respectively to S^0, S^2 or S^6. Each span_R{1,I} is an associative copy of C, even in O. Artin's theorem says any TWO elements of an alternative algebra generate an associative subalgebra. For a complex holomorphic function h satisfying h(conj z)=conj h(z), put

    h_A(x+yI) = alpha(x,y) + I beta(x,y),
    alpha+i beta = h(x+iy).

**[known]** At y=0 use its real value. This lift is slice regular where h is holomorphic. Zeta is slice meromorphic with its real pole at 1; xi is slice entire. This is slice regularity, not Fueter regularity. All products of intrinsic functions evaluated at the same q occur inside C_I, so no associativity ambiguity enters the zeta construction.

**[project; exact reduction]** For any such h, |h_A(x+yI)|=|h(x+iy)|. Every zero, norm bound, contour calculation in a slice, and real Taylor coefficient is therefore already encoded by TWO real coordinates (x,y). Orthogonal rotations R of the imaginary vector obey h_A(x+Rv)=Re h(x+i|v|)+(Rv/|v|)Im h(x+i|v|): as a vector-valued map, the symmetry is even larger than the algebra automorphism group. Aut(H)=SO(3), Aut(O)=G2; G2 is transitive on S6, but intrinsic radial equivariance does not require preserving octonionic multiplication.

**[known]** In Re q>1 the series sum n^{-q} converges absolutely, because

    n^{-q} = n^{-x}[cos(y log n) - I sin(y log n)],
    |n^{-q}|=n^{-x}.

The Euler product also holds there: all factors lie in the same associative commutative C_I. On Re q=1/2 each term has size n^{-1/2}, but the ordinary Dirichlet series is not convergent there; this term-size observation is not a construction of zeta on the line and not a cancellation proof.

**[project; numerical]** compute.py uses 60 decimal digits and deterministic seed 145. It implements quaternion/octonion multiplication by Cayley–Dickson recursion. Independent 200-term algebra-valued Dirichlet sums at 3+2.1I agree with the slice evaluation within the elementary absolute tail bound 1/(2*200²)=0.0000125. Tests in other sections that use the slice formula are checks of that formula's consequences, not independent discoveries of higher-dimensional zeros. No interval arithmetic is used.

## 2. Where 1/2 goes, and what the mirror really says

**[known]** Define xi(s)=s(s-1) pi^{-s/2} Gamma(s/2) zeta(s)/2, with removable values xi(0)=xi(1)=1/2. Its functional equation and real type give xi(1-s)=xi(s) and xi(conj s)=conj xi(s). Hence for the reflection sigma(q)=1-conj q,

    xi_A(sigma q) = conj(xi_A(q)).

The fixed set is Re q=1/2: a real line in C, a 3-dimensional affine hyperplane in H and a 7-dimensional affine hyperplane in O. Xi is real-valued there. Zeta is not generally real there; its gamma/phase factors must be retained or removed correctly.

**[project]** Thus RH has exactly equivalent geometric forms: every nontrivial zeta zero sphere has centre 1/2 on the real axis; every xi zero lies in that critical hyperplane. A zero a+ib with b>0 becomes a+bS^{d-2}. An off-line complex quartet becomes two spheres centred at a and 1-a. The trivial zeros of zeta stay isolated real points -2,-4,...; xi removes them. There is no special horn-like interval attached to 1/2.

**[project; numerical]** At the first three positive zeta heights 14.134725141734693790..., 21.022039638771554992..., 25.010857580145688763..., we sampled 12 directions per algebra, 36 points each. Maximum zeta norm was 4.211e-60 in both dimensions. Xi's unwanted imaginary norm on sampled critical spheres was below 3.219e-62; sampled mirror residuals were below 2.35e-62. These are finite precision residuals, not proof that all zeros lie there. They share the same underlying complex evaluations by construction.

## 3. The square folds entire critical spheres to single numbers

**[known/project; elementary proof]** Let u=q-1/2=a+v with v pure imaginary. Alternativity/power associativity gives

    u² = a²-|v|²+2av.

This is a negative real number precisely when a=0 and v!=0. Thus squaring folds the critical hyperplane minus its centre onto the negative real ray, and ONLY those points map onto that ray. All imaginary directions at radius r are identified with -r². On the real axis, squaring gives nonnegative numbers instead.

**[known]** Since xi(1/2+u) is even, define the entire scalar function E(w)=sum c_n w^n by E(u²)=xi(1/2+u). The same identity holds in H and O. E has order 1/2 and genus zero. RH iff every zero of E is negative real. The squared-coordinate method therefore survives unchanged, and does not acquire extra constraints from the fibre sphere.

**[project; numerical]** The first three folded zeros are -199.790454832386859..., -441.926150574082490..., and -625.542996894331058.... The normalized coefficients c_n/c_0 begin

    1,
    0.0231049931154189707889338104303,
    0.000248334053789144175723856445209,
    0.00000167435262798494986098862944241.

## 4. Positivity, moments, Jensen and Li: which tests survive?

**[known; Schoenberg/AESW–Edrei]** For E of this order with positive constant coefficient, RH is equivalent to its coefficient sequence being PF-infinity: every minor of its one-sided Toeplitz matrix is nonnegative. With T_ij=c_{i-j}, c_n=0 for n<0, this is an ALL-ORDERS condition, not just positive c_n. The genus-zero product under RH is E(w)=c_0 product(1+w/gamma_j²), with multiplicities. Conversely, the PF-infinity classification and the entire order bound force negative real zeros. This is the same scalar criterion in H and O, not a proposed quaternion/octonion determinant.

**[project; numerical]** For normalized c_n, the shifted contiguous minors det[c_{2+i-j}] for sizes 2,3,4 are respectively 2.29838963289736e-8, 1.19693056544332e-12, 4.23264440485044e-17. They are finite spot checks only. Their small sizes cannot establish PF-infinity. The earlier mixture counterexample and its negative fifth-order minor remain counterexamples in BOTH larger algebras.

**[known/project]** The positive theta kernel makes its Hankel moment matrices positive by the usual integral-of-a-square argument; that positivity is automatic and does not establish real-rootedness. Jensen polynomial hyperbolicity and Li's coefficient nonnegativity are likewise unchanged scalar RH criteria. The larger sphere supplies neither additional inequalities nor a new positive measure. There is no natural order relation on all quaternions or octonions that makes a non-real number “positive.” These tests live in their real coefficients, not in an invented ordering of H/O.

## 5. A useful derivative calculation: the angular directions collapse

**[project; derived here, standard slice-derivative structure]** At q=x+yI, y>0, decompose the tangent space into the 2-dimensional slice plane (dx,dy) and the d-2 directions tangent to the imaginary sphere. On the slice the derivative is complex multiplication by h'(z). On the angular subspace it is scalar multiplication beta/y, since the output imaginary vector is beta v/y. Thus

    det_R D h_A(q) = |h'(x+iy)|² [beta(x,y)/y]^{d-2},  d=4,8.

Both exponents are even, so this determinant is nonnegative. This also gives positive local indices at regular preimages of a non-real target. The derivation does not require an octonionic determinant.

**[project]** On the ENTIRE critical hyperplane xi_A is real, so beta=0 and all angular derivatives vanish, whether xi is zero or not. The derivative has rank at most 2 there (generically exactly 2 for y>0). Consequently “find where the full Jacobian vanishes” is not a zero detector: it finds a whole family of nonzero points too. This is a useful new-to-project obstruction to an attractive geometric shortcut, not a new theorem about RH.

**[project; numerical]** Central finite differences with step 1e-12 at x=.27,y=3.4 and random I reproduce the determinant formula with relative errors 4.03e-26 in H and 1.26e-25 in O. The determinant magnitudes are ~6.312e-8 and ~1.814e-17. Their difference reflects dimension and angular scaling, not different zeta information.

## 6. Loops, tubes and degree: make the counting statement precise

**[known]** A standard round S^{d-2} in R^d has a codimension-two meridian: a loop going around its small normal disk. The complement has fundamental group Z, for d=4,8. This assertion is about the standard unknotted sphere, not every embedded S2 in R4. This is an ordinary tame linking phenomenon, unlike the wild embedding of Alexander's horned sphere in R3. The intrinsic surface type alone never determines a complement.

**[known/project]** Although a meridian cannot contract in the DOMAIN complement, its image under h_A lands in A\{0}, which retracts to S^{d-1}; pi_1(S3)=pi_1(S7)=0. Therefore ordinary quaternionic/octonionic value-loop winding cannot carry the complex argument principle. Thicken the meridian over all imaginary directions instead: the boundary of a small normal tube is S^{d-2} x S1, a closed oriented (d-1)-manifold.

**[project; proof]** Near a conjugate pair of complex zeros of multiplicity m, the intrinsic factor is [(q-a)²+b²]^m times an intrinsic nonzero factor. In the normal plane, its value has complex winding m. The normalized tube map is homotopic, with orientations chosen as boundary orientations, to

    (I,phi) -> cos(m phi) + I sin(m phi).

Its signed degree is

    integral_0^{2pi} m sin^{d-2}(m phi) dphi
    ------------------------------------------------ = 2m.
            integral_0^pi sin^{d-2}(theta) dtheta

The two preimages for m=1 have the same sign: reversing the circle and taking the antipode of an even-dimensional sphere each reverse orientation. A real zero of complex multiplicity m has local degree m. A real pole of order m contributes -m; inverse q -> q^{-1} restricts to conjugation on S^{d-1}, of degree -1 in these even dimensions. For a meromorphic map this is a punctured-domain boundary-degree identity, not a Brouwer degree of a function defined at its poles.

**[known; EN44, GP09]** This is consistent with the degree-n quaternionic polynomial theorem of Eilenberg–Niven and the quaternionic/octonionic multiplicity theory. Eilenberg–Niven is a quaternionic paper; it should not alone be cited as the source of an octonionic theorem. Ghiloni–Perotti supply the octonionic zero/multiplicity framework, citing Jou (1950) and later proofs.

**[project; numerical]** For p(q)=q²+1 and regular target .3+.4e1, there are two explicitly solved preimages. Numerical d-by-d Jacobians are positive at both: .685135660952812... each in H, .0309240508994470... each in O. Thus signed count=2 independently in both dimensions. Direct one-dimensional integration of the tube model gives 2 for k=0,2,6. A 128-point circle of radius .01 around the first zeta zero has sampled winding 1, minimum sampled |zeta|=.0078989005. This is reduced-dimensional numerical evidence plus an exact degree derivation, NOT a high-dimensional mesh certification of xi.

**[project; corrected count]** On the boundary of a ball centred at 1/2, chosen to avoid xi zeros, the degree D is the full complex zero count in the corresponding disk, i.e. twice the upper-half-plane count. Let n_on count critical positive-height zeros in that disk WITH THEIR COMPLEX MULTIPLICITIES. RH in that disk iff D=2 n_on. Replacing n_on by raw sign changes is only sufficient for certification when equality is established: sign changes detect odd-multiplicity zeros but miss even multiplicities and cannot tell an odd multiplicity 1 from 3. RH alone does not imply D=twice the raw sign-change count. This is related to the logic of complete numerical verification, not literally Turing's full method: Turing supplies analytic control of missed zeros, not just this topological identity.

**[project; additional finite count check]** Sampling Hardy Z every .1 from 0 to 30 finds three sign-change brackets, [14.1,14.2], [21.0,21.1], [25.0,25.1]. A separate 512-sample xi contour around the disk |s-1/2|=30 has winding 6; mpmath's nzeros(30) returns 3 in the upper half-plane. The corresponding intrinsic boundary degree is 6 in both H and O. This example has matching counts, but all three numerical procedures remain non-interval-certified.

## 7. Heat flow: the same collisions, not a 4D or 8D diffusion

**[known; dB50, RT18]** In Rodgers–Tao's convention

    H_t(z) = integral_0^infinity exp(t u²) Phi(u) cos(zu) du,
    H_0(z) = xi(1/2+iz/2)/8,
    partial_t H_t = -partial_z² H_t.

There is a finite Newman constant Lambda such that all H_t zeros are real iff t>=Lambda. Rodgers–Tao proved Lambda>=0; RH is equivalent to Lambda<=0, hence, given that theorem, to Lambda=0. This is known mathematics, not an output of these experiments. No claim about the latest upper bound is needed here.

**[project; normalization audit]** To keep the critical plane at Re s=1/2 define

    K_t(s)=8 H_t(-2i(s-1/2)).

K_t is even about 1/2 and of real type, so has an intrinsic lift. It obeys partial_t K_t=(1/4)partial_s² K_t, not the Euclidean heat equation in R4 or R8. A real H_t zero z=r corresponds to an on-plane sphere of radius |r|/2; a non-real pair r±i eta corresponds to mirrored spheres at Re s=1/2±eta/2. This factor of two is important.

**[project; exact warning]** For the simple intrinsic polynomial h(q)=q², the Euclidean Laplacian is Delta_Rd h=-2(d-2): -4 in H and -12 in O. Holomorphic slice harmonicity therefore must not be confused with ambient harmonicity, and replacing the one-dimensional slice operator by a 4D/8D Laplacian would be a DIFFERENT deformation with no inherited Newman theorem.

**[project; exact toy + numerics]** The earlier toy h_t(z)=e^t cos z+(1/4)e^{4t}cos2z has threshold lambda=log4/3=.46209812037329687.... Its non-real roots just below threshold become real just above it. At t=lambda-.001 the roots are pi±.04475865257356325 i; at t=lambda+.001 they are pi±.04468411696215690. In the toy's unscaled placement s=1/2+iz (NOT the RT factor-1/2 placement), these are two off-plane sphere centres separated by .0895173051 below threshold, versus two critical radii separated by .0893682339 above. All checked root residuals are below 5e-61. Each simple sphere contributes 2: the degree stays 4 as 2+2 become a double sphere of degree 4 and split again.

**[project; explicit topological non-obstruction]** An especially transparent symmetric family is

    P_a(q)=[(q-1/2)²+1]²-a,  |a|<1.

For a>0 it has two critical spheres with radii sqrt(1±sqrt(a)); for a=0 one double critical sphere; for a<0 its squared-coordinate zeros -1±i sqrt(-a) give a pair of off-plane spheres. All cases have the same total degree 4 and obey the xi mirror symmetry; all folded polynomial coefficients 1-a,2,1 are positive. Thus ordinary degree, that mirror, and coefficient positivity do NOT force on-plane zeros, in either H or O. This does NOT rule out every refined equivariant invariant or analytic argument; it rules out these stated ingredients as sufficient. If an equivariant theorem is proposed, it must be checked against this family.

## 8. Mertens, reciprocal zeta and Riesz do not gain directions

**[known/project]** In Re q>1,

    zeta_A(q)^{-1}=sum mu(n)n^{-q}
                  =q integral_1^infinity M(x)x^{-q-1} dx.

All factors lie in C_I. M(x) is the SAME real integer-valued arithmetic function, not a function counting new quaternionic primes. Littlewood's criterion RH iff M(x)=O(x^{1/2+eps}) for every eps>0 is unchanged. The absolute norm of the integral is bounded in terms of Re q and |q| just as in a complex slice; the angular directions add no cancellation. The norm identity shows uniform-in-I bounds at fixed x,y are exactly the complex bounds. One must not simply assert absolute convergence of the original Mobius Dirichlet series in Re q>1/2 without the relevant RH assumption and convergence argument.

**[project; numerical]** At s=3+2.1i a 2000-term Mobius inverse sum agrees with 1/zeta(s) within its absolute tail bound 1/(2*2000²)=1.25e-7. results.json records the actual error and M(2000). This is only a finite arithmetic check.

**[known; R16]** The entire Riesz function

    R(x)=sum_{k>=1} (-1)^{k-1} x^k / [(k-1)! zeta(2k)]
        =x sum_{n>=1} mu(n)n^{-2} exp(-x/n²)

has real coefficients and hence an intrinsic H/O lift. RH iff R(x)=O(x^{1/4+eps}) on the POSITIVE REAL ray for every eps>0. This is not a bound over all quaternion or octonion directions.

**[project; exact negative-ray obstruction]** At x>0,

    R(-x)=-x e^x + O(x e^{x/4}),

because n=1 gives the first term and the absolute sum over n>=2 is at most x e^{x/4}sum_{n>=2}n^{-2}. Therefore no polynomial growth estimate over all of H/O can replace the Riesz ray criterion: it is already false on the negative real axis, independently of RH. Numerical ratios R(-x)/(-x e^x) tend rapidly to 1 for x=2,5,10,20; see data. Adding imaginary directions does not license rotating the criterion to other rays.

## 9. The non-real character experiment: spheres really become points

**[known/project; convention matters]** Fix one imaginary unit J. Take the primitive mod-5 complex character with chi(1)=1, chi(2)=J, chi(3)=-J, chi(4)=-1, chi(0)=0. Define the right-coefficient lift L_A(q,chi)=sum n^{-q}chi(n) in Re q>1, then extend slice regularly. The simple intrinsic formula alpha+I beta with REAL alpha,beta no longer applies. Instead, from U=L(x+yJ,chi), V=L(x-yJ,chi), define A=(U+V)/2 and B=J(V-U)/2, both in C_J. Then

    L_A(x+yI,chi)=A+I B
                =(1-IJ)U/2+(1+IJ)V/2.

Since I and J generate an associative subalgebra, this representation is safe even in O.

**[project; exact angular profile]** If U=0 and V!=0, then

    |L_A(x+yI,chi)|=|V| sqrt[(1-<I,J>)/2].

It vanishes ONLY at I=J; the conjugacy sphere contains one isolated zero, not a spherical zero. If U and V both vanish, the entire sphere vanishes instead. More generally coefficients in one C_J imply all nonspherical zeros lie in that coefficient slice: away from it the components in C_J and its orthogonal part force both real-stem components to vanish for a zero. This is a direct representation-formula deduction, consistent with GP09's zero structure theorem.

**[project; numerical]** mpmath finds an L(s,chi) zero at

    rho=.5+6.18357819545085391437751730970869 J,
    |L(rho,chi)|=1.70e-65,
    L(conj rho,chi)=2.26997224075759534-1.66861657054454230 J.

At the same x,y, the lift norm is ~0 along J, 2.81727798296616231 along -J, and 1.99211636624293209 along a direction perpendicular to J. Both H and O reproduce these values. “0” in some output samples is working-precision cancellation; the complex root residual is recorded separately. This confirms an actual point-versus-sphere distinction, not a claimed off-critical zero of a Dirichlet L-function. The zero lies on the generalized critical line.

**[known/project]** Slice conjugation conjugates the coefficients: f^c=L_A(q,conj chi). The normal/symmetrized function f*f^c has real coefficients; on C_J it equals L(s,chi)L(s,conj chi), and its zeros circularize the zeros of either character. It is NOT the pointwise norm |L(q)|² away from the real axis. At the above root, the normal product residual is 4.79e-65; the conjugate character at the same positive-height root is nonzero, 2.26997224075759534+1.66861657054454230 J. Symmetrization therefore supplies a full sphere even though the original L lift vanishes at only one point of it.

**[project; ceiling]** Because all chi values lie in one complex coefficient plane, the experiment still contains no irreducibly octonionic arithmetic. Its angular profile is exactly reconstructible from the two complex values U,V. Breaking zeta's real symmetry makes the geometry different, but not its arithmetic information richer than the original complex L-function and its conjugate.

## 10. A genuinely octonionic effect, with a controlled example

**[known; GP09 Remark 3]** Choose quaternionic units i,j and an octonionic unit k outside their quaternionic subalgebra. For the right-coefficient slice polynomial f(q)=q i-j and constant g(q)=k, the slice product is

    (f*g)(q)=q(ik)-jk,

not generally the pointwise (q i-j)k. The former has its unique zero at -ij, whereas f has its zero at ij. Multiplying by a nonzero constant on the right has shifted a zero because reassociation is illegal: this cannot happen for pointwise multiplication, whose norm is multiplicative.

**[project; numerical/exact arithmetic check]** With Cayley–Dickson basis i=e1,j=e2,k=e4, compute.py checks f(ij)=0, (f*g)(-ij)=0, |(f*g)(ij)|=2, and |f(ij)k|=0 exactly at the chosen working precision (integer basis arithmetic). Both locations have real part 0 and imaginary radius 1. The angular zero changes but its conjugacy sphere does not; GP09 proves the corresponding normal-function multiplicativity and conjugacy-class preservation.

**[speculative]** A future construction involving coefficients outside any common associative subalgebra could have genuinely octonionic angular geometry. But zeta provides no canonical such coefficients. One would first need an independently motivated arithmetic object, multiplication convention and well-posed functional equation. Assigning random i,j,k weights would make a different function, not a route to RH. Even the published camshaft example preserves the two coordinates that determine criticality.

## 11. Finite completeness and what remains open

**[project]** Complete at the dictionary level: every requested existing tool has a precise H/O interpretation, and we have an intrinsic reduction theorem, tube-degree derivation, derivative determinant calculation, heat normalization audit, exact non-obstruction family, Riesz negative-ray exclusion, complex-character experiment and true octonionic counterexample to careless multiplication. None proves RH or introduces a new field.

**[project]** Not done: interval-certified zeros, high-dimensional triangulated degree integration, global positivity, an equivariant-degree classification, and a canonical nonassociative arithmetic L-function. Numerical sampling cannot substitute for these. The toy collision is not a computation of the actual Newman constant.

**[speculative; bounded follow-up]** A useful next session is to audit one symmetry-sensitive candidate invariant against P_a BEFORE touching xi; or parameterize the mod-5 angular profile at several zeros and interval-check the isolated-versus-spherical classification. Reject any candidate quantity that is algebraically a function only of x,|Im q| unless its real/complex counterpart is itself useful.

## Sources and reproducibility

**[known; primary references]**
- GP11: R. Ghiloni and A. Perotti, *Slice regular functions on real alternative algebras*, Advances in Mathematics 226 (2011), 1662–1691. https://doi.org/10.1016/j.aim.2010.08.015 . General alternative-algebra construction. Search metadata inspected; full article not read in this pass.
- GP09: R. Ghiloni and A. Perotti, *Zeros of regular functions of quaternionic and octonionic variable: a division lemma and the camshaft effect* (2009 preprint; Annali di Matematica Pura ed Applicata, DOI https://doi.org/10.1007/s10231-010-0162-1). Full author preprint read in this pass: https://arxiv.org/html/0904.2667 . In particular Corollaries 2 and 5, Theorem 6, Lemma 7, Remark 3. It states and uses Artin's two-generator theorem.
- GSS: G. Gentili, C. Stoppato, D. C. Struppa, *Regular Functions of a Quaternionic Variable*, Springer, 2013. https://doi.org/10.1007/978-3-642-33871-7 . Standard reference; not freshly read. Earlier foundational papers are cited in GP09.
- EN44: S. Eilenberg and I. Niven, *The “fundamental theorem of algebra” for quaternions*, Bulletin AMS 50 (1944), 246–248. https://doi.org/10.1090/S0002-9904-1944-08125-1 . Search excerpt gives degree theorem; full DOI fetch timed out, not represented as freshly read.
- Turing: A. M. Turing, *Some calculations of the Riemann zeta-function*, Proc. LMS (3) 3 (1953), 99–117. https://doi.org/10.1112/plms/s3-3.1.99 . Cited for verification methodology, not identity of the simplified topological count with the whole method.
- Schoenberg: I. J. Schoenberg, *On Pólya frequency functions. I. The totally positive functions and their Laplace transforms*, J. Analyse Math. 1 (1951), 331–374. AESW: M. Aissen, A. Edrei, I. J. Schoenberg, A. Whitney, *On the generating functions of totally positive sequences*, PNAS 37 (1951), 303–307. https://doi.org/10.1073/pnas.37.5.303 . DOI fetch timed out. Applied through the explicitly stated entire genus-zero specialization, not a newly established classification.
- dB50: N. G. de Bruijn, *The roots of trigonometric integrals*, Duke Math. J. 17 (1950), 197–226. https://doi.org/10.1215/S0012-7094-50-01720-0 . Prior project scan and notes: ../squared-coordinate/debruijn1950.pdf and scale-heat-bridge.md. Cite standard theorem, not a fresh rereading of the scan today.
- RT18: B. Rodgers and T. Tao, *The de Bruijn–Newman constant is non-negative*, preprint 2018; Forum of Mathematics, Pi 8 (2020), e6. Full introduction and normalization read: https://arxiv.org/html/1801.05914 . https://doi.org/10.1017/fmp.2020.6 . Normalization in equations (1)–(4).
- R16: M. Riesz, *Sur l'hypothèse de Riemann*, Acta Math. 40 (1916), 185–190. https://doi.org/10.1007/BF02418544 . Project Euclid bibliographic page opened, body not provided by fetch. Formula/criterion carried forward from the existing Riesz page and checked here on its stated ray.
- J. E. Littlewood, *Quelques conséquences de l'hypothèse que la fonction zeta(s) de Riemann n'a pas de zéros dans le demi-plan Re(s)>1/2*, CR Acad. Sci. Paris 154 (1912), 263–266. Prior project Mertens reference, not freshly read.
- X.-J. Li, *The positivity of a sequence of numbers and the Riemann hypothesis*, J. Number Theory 65 (1997), 325–333. https://doi.org/10.1006/jnth.1997.2137 . Cited only for the unchanged classical criterion.
- M. Griffin, K. Ono, L. Rolen, D. Zagier, *Jensen polynomials for the Riemann zeta function and other sequences*, PNAS 116 (2019), 11103–11110. https://arxiv.org/abs/1902.07321 . Prior project reference; eventual fixed-degree hyperbolicity must not be mistaken for all degrees/shifts.

**[project; files]** compute.py runs offline and writes results.json plus stdout (saved as run.log); assertions check tolerances. figures/ contains schematic SVGs, explicitly not projections of all eight coordinates. Numerical outputs carry decimal precision but not rigorous error enclosures. The polynomial Jacobian check tests multiplication/degree independently; the intrinsic zeta sphere samples are implementations of the exact complex reduction.


**Pointer (12 Nov 2026, ~8:30 AM PST):** binary tetrahedral 2T = SL(2,3) version of the Q8 quaternion-Frobenius work — `tetrahedral-frobenius-20261112.md` (data/code `tetra1112/`, figure `figures/tetrahedral-frobenius.png`). Reduced trace is the only choice-free linear statistic for every non-abelian G ⊂ H^× [project, proved]; 2T fixed-convention pure parts always involve the cubic twists ρ⊗ω [project, proved]; aligned 2T product has real exponents ½, ½, ⅛, a (s−½)^{3/8} central point and a sixth-root (not square-root) norm barrier at ½ [project, numerics]. No RH claim.

- 12 Nov 2026 ~8:50 PM PST [project, interval arithmetic]: the forced central zeros of both 2T Artin L-functions (2.26569.24t7.b.a, .a.a) are simple: L'(rho,1/2) in (2.62,2.85) and (2.97,3.07). See tetra-central-simplicity-20261112.md. Page #binary-tetrahedral-exponents, version 2ed2cec9.

- 13 Nov 2026 ~8:00 AM PST [project, float numerics; known theorem]: zero side of the 2T aligned pure part rebuilt from PARI zeros (u-direction: ζ, L(ω), L(ω̄), L(ψ3); n-direction: asymmetry of zeros of L(ρ⊗ω), needs an odd test function; central zero of L(ρ) contributes 0). Residuals ≤ 10⁻¹⁰ (T=1), ≤ 10⁻¹⁴ otherwise. See tetra-pure-zero-side-20261113.md, tetra-zero1113/, figures/tetra-pure-zero-side.png.

- 13 Nov 2026 ~5:05 PM PST [project, exact identity + local root numbers + interval/float numerics]: binary icosahedral 2I aligned product F₁₀ along a 10-fold axis = L(ρ)^{1/3} L(ρ′)^{−1/6} L(Sym³ρ)^{1/6} · ζ(2s)^{1/8}·(A₅ L-functions at 2s, real ∓1/8) · phases · H (H holomorphic for Re s > 1/3). In the galpol field (ramified 2, 11, 23), W(ρ) = W(ρ′) = +1 but W(Sym³ρ) = −1, so the central exponent is 1/6 − 1/8 = 1/24 (conditional on Artin's conjecture for these icosahedral L-functions and on L′(Sym³ρ, ½) ≈ 14.97 ≠ 0, a float). Measured slope 0.041647. See icosa-exponent-20261113.md, icosa-exp1113/, figures/icosahedral-exponent-site.svg. No RH claim.

- 13 Nov 2026 ~9:40 PM PST [project, exact character theory + float numerics; known theorem]: zero side of the 2O Model A quaternion prime sum (galpol [48,28]). Real part = ½·L(ρ) exactly; the pure part (4-, 3- and 2-fold axes) at k = 1 is built only from S₄ L-functions (ζ, L(ε), L(χ₂), L(ψ₃), L(ψ₃⊗ε)), with weight 0 on L(ρ), L(ρ⊗ε), L(σ₄); faithful L-functions enter only through prime squares. All 2O characters are real, so there is no up/down zero-asymmetry channel as in 2T (also true for Q8, 2I). Residuals ≤ 4·10⁻¹⁰ (T = 1), ≤ 10⁻¹⁴ otherwise; PARI zeros uncertified. See octa-pure-zero-side-20261113.md, octa-zero1113/, figures/octa-pure-zero-side.png. No RH claim.


- 14 Nov 2026 ~7:50 AM PST [project, float numerics + proof from known local facts]: two icosahedral (2I) Artin L-functions with root number −1 and forced central zeros: LMFDB 2.3806401.120.a (p = 1951, totally real, Γ_ℝ(s)², checkfeq 2⁻⁵⁵, L′(½) ≈ 14.640) and 2.4583881.120.a (p = 2141, complex conjugation = −1, Γ_ℝ(s+1)², 2⁻⁵⁴, L′(½) ≈ 8.388). 5A/5B ambiguity resolved by an oriented quintic discriminant. Local rule: where D_v is abelian, W_v = −1 exactly when the local Artin symbol of −1 is the quaternion −1 (inertia C₁₀ at 1951 with 195 odd; complex conjugation at ∞ for 2141). Corrects two 13 Nov claims (p¹⁶ for 1951; "+1 control" for 2141). Conditional on holomorphy (Artin's conjecture). See icosa-oddW-20261114.md. No RH claim.


- 14 Nov 2026 ~12:45 PM PST [project, interval arithmetic (Arb); conditional on Artin holomorphy]: the forced central zeros of the two 2I root-number −1 fields are **simple**. L′(ρ, ½) ∈ 14.6404198415908494271272808 ± 4.9·10⁻²⁶ (2.3806401.120.a, p = 1951, Γ_ℝ(s)²) and 8.38821569820444695438668613 ± 3.7·10⁻²⁷ (2.4583881.120.a, p = 2141, Γ_ℝ(s+1)²); the √5-conjugates ρ′ also have W = −1 and L′(ρ′, ½) ∈ 3.7050867341305652169151417 ± 5.2·10⁻²⁶, 6.188781635851670992970410311 ± 8.9·10⁻²⁸. Root numbers certified (under the same hypothesis) by θ(½)/(2θ(2)) = −1 ± 10⁻³³. Two runs (cutoff 60/300 bits, 80/400 bits) agree; PARI floats within 8·10⁻¹⁶; a_p audited against the LMFDB quintic (2 258 primes per field, 0 mismatches). No RH claim. Note `icosa-oddW-certificate-20261114.md`, scripts `icosa-cert1114/`, figure `figures/icosahedral-odd-sign-site.svg`.

- 14 Nov 2026 ~2:00 PM PST [project, proved; unconditional — Chebotarev, Jordan's lemma and Artin L-functions on Re s = 1 only, no Artin conjecture]: **uniform edge bound.** For every finite G ⊂ H^× and axis I, the Model A correction H (after peeling the s and 2s layers) has edge σ_abs = 1/k₀, k₀ = first ghost layer Φ_d ≢ 0 with d ≥ 3; it is also the true holomorphy edge when the trivial weight c₁ ∉ ℤ_{≤0}. Standard (longest-cyclic) axis: σ = 1/3 for Q8, Dic_n (n ≥ 4), 2O, 2I and σ = 1/4 for Dic₃, 2T; cyclic groups on their own axis have H ≡ 1. So **1/6 ≤ ½ − σ ≤ 1/4, sharp, attained**. Over all axes: 1/max(4, P(|G|)) ≤ σ ≤ 1/3; Dic_p on a 4-fold axis gives σ = 1/p, so the distance approaches ½ but never attains it. No RH claim. See uniform-edge-20261114.md, edge1114/, figures/uniform-edge-site.svg.
