Zeta Zeros
Display
Theme
Motion

A long tangent · downward transfer ledger

What bigger spaces can tell the number line

Bigger spaces can hold ordinary zeta inside them. Sometimes a theorem in the bigger world really does come back. Sometimes it is a beautiful analogy. Here is the return ticket to ask for.

[project] This is a map of implications, not progress on RH. [known] means established mathematics; [project] means checked or organized here, not a novelty claim; [speculative] means a missing theorem is still missing. Precise hypotheses, primary sources and calculation limits are in the full ledger. There is also a plain-English primer.

A return ticket, not just another dimension

[known] Ordinary RH says every nontrivial zero of ζ(s) sits halfway across its critical strip. Those zeros influence the fluctuations of primes on the positive number line. A theorem about a different function needs a connecting theorem before it can constrain these zeros—and, through them, the primes.

[project] The useful question is not “have we run out of bigger spaces?” It is “what exact bridge carries this particular fact back?” Number-field degree, geometric dimension, representation size and quaternionic directions are different kinds of bigger.

Figure 1

Three solid bridges lead to ordinary zeta: an actual factor with entire quotient, an exact RH criterion, and a locally uniform zero-free limit. More directions, another solved RH, and matching statistics have only dotted connections.
[project] The solid arrows are logical mechanisms. Dotted arrows need an additional identity or theorem; resemblance alone is not a transfer.

Containment

[known] F=ζ×G, with G holomorphic. An ordinary zero cannot disappear. A zero-free theorem for F therefore constrains ζ.

Exact reformulation

[known] Prove all the required positivity or approximation conditions, and a known equivalence returns RH.

A controlled limit

[known] Locally uniform zero-free approximants have a zero-free limit unless that limit vanishes identically: Hurwitz's theorem.

The larger field really contains ordinary zeta

[known] Our Q₈ field is a genuine eight-degree extension of the rationals. Its zeta factors as

ζK(s) = ζ(s) L(s,χ₅) L(s,χ₄₁) L(s,χ₂₀₅) L(s,ρ)².

[known] The Aramata–Brauer theorem makes ζK/ζ entire for a Galois extension. No pole can cancel an ordinary-zeta zero. RH for the whole field function would imply ordinary RH. So would a suitable zero-free region. The general non-Galois quotient requires care; the Galois hypothesis is doing work here.

[project] We rebuilt 2.42025.8t5.a.a with PARI, matched 5000 arithmetic coefficients exactly, and listed the factors' positive zeros below height 30. Three ordinary zeros, 11 + 21 + 28 quadratic zeros, and 55 Artin heights counted twice give 173 field zeros with multiplicity.

Figure 2

Five factor rows and a combined field row show zero heights from 0 to 30. The ordinary row has 3; quadratic discriminants 5, 41, 205 have 11, 21, 28; the Artin row has 55, each contributing twice. The combined positive count is 173; the centre has multiplicity 2.
[project] The green Artin contribution is doubled. The field union is exactly what the factor theorem predicts; the original lists and sampled winding were numerical. New interval contours certify all ordinary/quadratic factors; a complete fresh rerun reproduces the earlier Artin enclosure.
30.0

At height 30: 3 + 11 + 21 + 28 + 2×55 = 173 positive zeros with multiplicity.

[project] Two sampled contours agree at winding 348 = 2×173+2, including both signs of the height and the double centre. PARI internally factors its Dedekind evaluator: this is not an independent direct computation of the field zeta, and a sampled contour is not a certified enclosure.

[project] A stronger second check: 160-bit interval contours certify total counts 6, 22, 42 and 56 for ordinary ζ and the three quadratics in the rectangle −½≤Re(s)≤1½, |Im(s)|≤30. Together with the freshly rerun Artin certificate (221.824 seconds, identical non-timing output), the factor theorem certifies the combined count 348. This is a factored certificate, not a new direct field contour. Proof, bounds and dependency scope · Combined record.

[known] The arrow is one-way. The Artin factor is forced to vanish at ½; the field function therefore vanishes there twice. Ordinary ζ(½)≈−1.46035 is not zero. A larger zero need not be a smaller zero.

Tensoring gives another honest bridge

[known] For this Q₈ representation, ρ⊗ρ* splits as the trivial character plus the three quadratic characters. Its L-function contains ζ. Odd tensor powers just repeat ρ; even powers contain ζ. The symmetric square has the three nontrivial quadratics but no trivial factor. The symmetric fourth power, by contrast, contains two trivial characters: its L-function contains ζ². We checked tensor powers through 12 and symmetric powers through 24 exactly.

[known] Tensoring combines prime-by-prime arithmetic eigenvalues—not old zero locations. Knowing every zero of L(s,ρ) would not by itself give the tensor L-function's zeros. Gelbart–Jacquet and Yang supply adjoint holomorphy in stated small-rank automorphic cases, allowing similar Rankin–Selberg-to-ζ transfers; general rank must retain the holomorphy gate.

Where RH really is a theorem

[known] Over a finite field, counting points on a smooth projective curve produces a different zeta function. Hasse and Weil proved its zeros lie on a circle. Frobenius is a real geometric operator, and intersection inequalities on a product of curves control it. That is additional structure, not just additional axes.

[project] Take y²=x³+x+1 over the field with five elements. It has 9 points including infinity. Its zeta numerator is 1+3T+5T², with zeros −0.3±0.331662479i. Both have radius exactly 1/√5.

Figure 3

The elliptic curve over F5 has two zeros on a circle of radius 1 over square root 5. Its product surface has zeros on two circles: weights 1 and 3, at radii 5 to the power minus one half and minus three halves.
[project] We checked extension-field point counts exactly through degree 8. The surface E×E shows why higher-dimensional RH has separate cohomological weights, not one universal circle.

Turn Frobenius; keep its proven radius

[known / project illustration] After dividing each Frobenius eigenvalue by √5, both lie on the unit circle. Extending the field to 5ⁿ raises them to the nth power: their angles multiply while the normalized radius stays 1.

normalized radius 1
1

Over F₅: 9 curve points; 81 points on E×E.

[known] Deligne proved the finite-field theorem in every dimension. Cohomology separates the weights; the trace formula identifies Frobenius; geometric and monodromy methods prove the bounds. This is not a theorem about the global Euler product of an elliptic curve over the rationals. That Hasse–Weil RH is generally still open.

A spectral model, with a qualification

[known] For compact hyperbolic surfaces, the nontrivial spectral zeros of Selberg zeta satisfy s(1−s)=λ for Laplace eigenvalues λ. A self-adjoint Laplacian makes λ real. Values λ≥¼ give the halfway line, but small positive eigenvalues give real exceptional zeros off it. So “Selberg proves all zeros are on the line” would omit a genuine exception.

[speculative] Deninger's arithmetic cohomology/foliation program seeks a flow and a positive pairing. Connes–Consani's arithmetic/scaling sites seek another geometric route to the explicit formula. These are distinct programs with real component results, not completed ordinary-RH proofs. The missing positivity or full spectral identification is exactly the proposed bridge.

E₈: the mirror centre has no zeros at all

[known] The E₈ lattice zeta factors as 240·2−sζ(s)ζ(s−3). Its mirror is centred at 2. Yet the whole centre line is unconditionally zero-free, as is the open band between real parts 1 and 3. On the line the completed function is a negative constant times strictly positive factors.

Figure 4

E8 inherited zeros occupy strips near real parts zero to one and three to four. The centre band one to three is shaded as zero-free. A second panel shows minus the completed value along the centre line staying positive at all sampled heights.
[project] Under RH the inherited nonreal zero lines are ½ and 3½. The centre line 2 is zero-free by an exact identity, not merely by this numerical plot.

[known] So proving every nonreal E₈ zero lies on ½ or 3½ would imply ordinary RH. But mirror symmetry alone does not do it. At the centre, the completed value is −5/6; the raw continued lattice zeta is −5π²/6. Rescaling the exponent cannot turn its empty centre into a zero line.

[known] The quaternion ideal-index zeta gives a different exact bridge: ζ(2s)ζ(2s−1) times finite factors. All its nonreal zeros lying on ¼, ½ or ¾ is equivalent to ordinary RH. This is not the same summation object as a four-dimensional lattice zeta.

[known] A genuinely higher-dimensional arithmetic scheme can be equally explicit: the zeta of projective n-space over the integers is ζ(s)ζ(s−1)…ζ(s−n). It contains ordinary ζ, but repeated shifts do not magically solve the original factor. General geometric zetas are alternating products and quotients, so possible cancellations must be checked.

A statistical match is not a zero-by-zero implication

[known] Higher-rank GL(n) L-functions and random-matrix models give rich zero statistics. Montgomery and Rudnick–Sarnak proved correlation results under restricted hypotheses and Fourier support. Katz–Sarnak distinguish family symmetry types and low-lying zeros. Unrestricted GUE for ordinary zeros is not thereby a theorem; bulk spacing and family symmetry are different questions.

Figure 5

Histogram of the first 500 zeros and a step curve for the first 5000 ordinary-zeta zeros, with the GUE sine-kernel curve and the uncorrelated density-one benchmark. Small separations are suppressed.
[project] Two completed low-height samples at 25 decimal digits: 1725 pairs from 500 zeros, 17492 pairs from 5000. The larger sample ends at height 5447.861998. Bin RMS discrepancy from GUE changes from .09349 to .03809. Descriptive only: no significance test and no RH implication.

[known] Even a perfect limiting bulk law can miss finitely many exceptional zeros. A universal claim about every real part needs a stronger bridge than an averaged histogram. Katz–Sarnak's author survey makes the theorem/conjecture distinctions explicit.

Bigger spaces with exact return tickets

Li coefficients

[known] Every coefficient nonnegative ⇔ RH. Twenty or 500 positive coefficients is not “every”. In our field, coefficients add across factors; a positive total at one index does not imply positive summands.

Li, 1997

Weil positivity

[known] The right explicit-formula pairing is positive on every admissible test square ⇔ RH. A finite positive matrix, or an unrelated squared norm, is not that universal theorem.

Precise pairing and local results

Nyman–Beurling

[known] A specified approximation problem in a Hilbert space has error tending to zero ⇔ RH. One finite error bound weakens as zero height grows; a genuine convergent sequence would reach every fixed zero.

Báez–Duarte's strengthening

de Bruijn–Newman flow

[known] An extra time parameter moves the zeros. All are real exactly above a threshold Λ. Rodgers–Tao proved Λ≥0; RH is now equivalent to Λ=0. A positive-time success need not run backwards.

Rodgers–Tao

Jensen polynomials

[known] The full real-rootedness criterion is equivalent to RH. Fixed-degree eventual real-rootedness is proved, but is not uniform control of every degree and shift. The quantifiers matter.

Griffin–Ono–Rolen–Zagier

A bigger family that already ruled out a whole line

[known] This is not all hypothetical. A classical inequality couples three shifted values of ordinary ζ: ζ(σ)³|ζ(σ+it)|⁴|ζ(σ+2it)|≥1 for σ>1. Its Euler-product terms are nonnegative because 3+4cos θ+cos 2θ=2(1+cos θ)².

[known] If ζ vanished at 1+it, the fourth-power zero would overpower the third-power pole as σ approached 1. The product would tend to zero, contradicting the lower bound. So Re(s)=1 is zero-free. The argument does not extend automatically inside the strip: that positive expansion was only justified to the right of 1. Complete proof and primary references.

[project] Twenty 192-bit interval examples illustrate the inequality, but the proof is the universal identity and limit, not those samples.

Two return tickets turn out to share a matrix

[known] Lagarias gives an exact identity: certain Weil pairings equal λₙ+λₘ−λ|n−m|, where λ denotes Li coefficients. Positivity of every such finite compression is one way to organize the infinite criterion—not a shortcut around it.

[project] Using 2048-bit interval arithmetic, we enclosed 50 positive Li coefficients and certified a positive 32×32 compression. Its last elimination pivot is only 1.408×10−110. Ordinary floating-point elimination had already produced a spurious negative pivot at index 7. Here precision is part of the evidence, and finite positivity still does not imply RH. Interval record · script.

A finite approximation that really constrains every zero

[known] In the tail-cancelled Nyman–Beurling formulation, an off-line zero β+it with β>½ prevents a chosen approximant's squared L² error from falling below (2β−1)/(β²+t²). The Mellin transform supplies the identity; Cauchy–Schwarz supplies the inequality. It is an actual quantitative return ticket.

[project] We computed exact-form Gram integrals with 192-bit Arb enclosures and chose explicit rational-decimal coefficients. With generators through a=2 the squared error is 1−log 2≈.306853; through a=30 it is .014556733577. These finite bounds are far weaker than known results about ζ zeros. No new zero-free region is claimed.

Figure 6

Squared approximation errors fall from .30685 for the generator a equals 2 to .01455673 using integers through 30. The adjacent panel states the exact off-line-zero obstruction and why finite error cannot settle all heights.
[project] The infinite Gram sums are interval enclosed using a periodic digamma formula, not a finite-tail guess. The experiment supplies a finite bound, not density of the full span or progress on RH.

[known] A sequence of rigorous errors tending to zero would exclude every fixed off-line zero. The infinite-dimensional statement is essential: the weakening at large height is visible in the denominator.

[project] Interval results and exact chosen coefficients · reproduction script.

What comes back from our six hypercomplex threads?

DifferenceDownward transfer?
Directional zeros[known / project] Not from direction alone. A separate character-product factorization may provide one.
Forced central zeros[known] They force a field zero, not an ordinary ζ zero at ½.
Bracket-moved zeros[project] Our octonion star-product toys have no established ordinary-ζ factor. No transfer yet.
Quarter-lines[known] Yes, through the scaled ζ factors and a no-cancellation check—not because quarters look special.
Holes at infinity[project] The present topological information does not constrain the zeros' real parts.
Centre n/4[known] It specifies a mirror, not a zero location. E₈ transfers through its product, while its centre stays empty.

[known] The intrinsic lift of ordinary ζ restricts exactly to ζ on each complex slice. “All its nontrivial zero spheres have real coordinate ½” is therefore equivalent to RH. Roundness and multiplicity counts alone do not prove that coordinate statement.

[speculative] A useful next search is position-sensitive: an operator, pairing, approximation theorem or factor identity whose ordinary slice is exactly a known ζ object. Prove its positivity or zero-location property independently, then use the bridge. This ledger marks today's inference limits, not the end of the territory.

And then, back to the actual number line

[known] If RH held for our larger Galois-field zeta, it would hold for ordinary ζ. That would imply prime-power counting error bounded by a constant times √x log²x, and prime-counting error bounded by a constant times √x log x, uniformly along the positive number line. Von Koch's original 1901 paper gives this last prime-counting bridge.

[project] A certified zero list through height 30 is not that all-height premise. A finite-range prime estimate would also need a controlled explicit-formula tail. The return ticket is real; the missing universal hypothesis is still missing.

Which return tickets actually carry weight?

[project] First: a factor that cannot be cancelled. Second: an exact criterion with all its quantifiers intact. Third: a certified extraction map from larger fields. Geometric analogies and family statistics can guide a search, but do not supply an ordinary-zero theorem by themselves.

Figure 7

Three ranked transfer mechanisms, with analogies explicitly requiring an additional theorem.
[project] Ranked by the logical bridge available, not by a prediction of an RH breakthrough.

[known] A zero-free region or zero-count upper bound for a larger Galois-field zeta descends to ordinary ζ. The converse does not follow. Known generic field bounds usually do not improve ζ’s direct estimates; our finite experiments do not do so.

Li positivity: finite and universal are different claims

[known] Completed-factor Li coefficients add. Finitely many positive field coefficients do not isolate the ordinary summand. But if every field Li coefficient is nonnegative, the Dedekind Li criterion gives field GRH; zero inclusion then gives ordinary RH. The all-index implication is valid, not a subtraction of positive numbers.

Figure 8

Twenty field Li coefficients decomposed into ordinary, quadratic and Artin parts; the first ordinary contribution is only 0.286 percent.
[project] Numerical PARI Taylor coefficients, not interval-certified by the zero counts. Additivity is exact; positivity of the displayed finite sample is not the universal Li premise.

Full ranked verdict, primary sources and missing gates · Bombieri–Lagarias: the all-index criterion.

Which factor owns the zero?

[known] Stark’s simple-zero lemma is a precise ownership rule: a simple field-zeta zero belongs to exactly one degree-one Artin factor. If ordinary ζ also vanishes there, it is that factor; the other irreducible factors are nonzero.

Figure 9

At three ordinary zeros the field order is one; at the centre ordinary zeta is nonzero while a simple Artin zero contributes twice to the field.
[project] Fresh interval local counts. These particular field evaluations used the factors, so this is not independent downward evidence; the next section supplies that different pipeline.

[project] The honest ceiling: a hypothetical off-line ordinary zero of multiplicity m can occur in every field with the same order M=m, while all extra factors stay nonzero. Restriction identities and the Heilbronn norm bound remain consistent. In Q₈, known factor holomorphy already makes that bound automatic.

Figure 10

The assignment M equals m and all extra orders zero satisfies the multiplicity budget even at a hypothetical off-line zero.
[project] A consistency model showing what the bookkeeping alone cannot exclude. It is not evidence that such a zero exists.

[speculative] A new independent location or upper-count theorem could rule the model out. Merely adding fields, directions or symmetry does not. Derivation, stronger solvable-group budgets, source limits and certificates.

A different computation really comes back down

[project] We then stopped evaluating ordinary ζ and character L-factors. Each proper subfield zeta was computed from its own prime-ideal splitting data, certified units and a Gaussian smoothed functional equation with interval tails. All 4000 checked ideal-count coefficients matched PARI.

Nζ(30) = (14 + 24 + 31 − 63) / 2 = 3.

Figure 11

Independent subfield counts 14,24,31,63 recover the three positive ordinary zeros through height30; local recovered orders0,1,1,1 match the references.
[project] Whole-strip counts, multiplicity included. Reference ordinary data were opened only after the independent certificate passed.

[known / project] The factor ½ is essential: twice the ordinary order equals the three quadratic-subfield orders minus the biquadratic order. The same identity recovers ordinary orders 0,1,1,1 at the centre and first three zeros. An independent complete count leaves room only for those three zeros in this finite window.

[project] “Independent” means a different computation, not different mathematical information. The coefficients still encode the same functions. This verifies a long-known finite result, not RH. The direct degree-eight field evaluator remains open and is unnecessary for this projector.

Formula, every error bound and reproduction instructions · Interval certificate · Separate verification.

The notebook behind the tangent

[project] Version downward-20261110-v1. The research keeps exact arithmetic, non-interval numerical checks, established theorems and speculative programs separate. No joint-field claim and no RH-progress claim.

[project] Continue with the hypercomplex exploration, zero statistics, or the ordinary RH page.