# What bigger spaces can tell the number line

A plain-English tangent for Grant. Version downward-20261110-v1.
Every factual paragraph is labelled: **[known]** established mathematics;
**[project]** checked or organized in these notes; **[speculative]** a possible
route without its missing proof. The full citations and hypotheses are in
[the ledger](ledger.md).

## The bigger setting is useful when it comes with a return ticket

**[project]** Your instinct leads to a productive question: not “are higher
spaces a dead end?”, but “which information comes back, and through what bridge?”
There are real bridges. We found clean ones in our own quaternion examples.
There are also wonderful higher-dimensional facts that teach us how a proof
could work without already being a proof about ordinary zeta.

**[known]** The ordinary Riemann Hypothesis concerns zeros of a function of a
complex number: they should all sit halfway across a certain strip. Those zeros
control fluctuations in the primes on the positive number line. Gathering more
zero information can therefore improve prime information—but a theorem about a
different function needs a connecting theorem before it can constrain our primes.

## 1. Put ordinary zeta inside a larger product

**[known]** This is the clearest return ticket. If the bigger function is ordinary
zeta multiplied by other functions with no poles, an ordinary-zeta zero cannot
disappear. It is also a zero of the bigger function. Prove that the bigger
function has no zeros in some region, and ordinary zeta has none there either.
The direction matters: a zero of the bigger function may belong only to a new
factor, not to ordinary zeta.

**[known]** Our eight-degree Q8 field does exactly this. Its field zeta is
ordinary zeta × three quadratic L-functions × the square of our quaternionic
Artin L-function. The classical Aramata–Brauer theorem makes the no-cancellation
bridge rigorous for a Galois field extension. It does not automatically cover
every arbitrary number-field extension.

**[project]** Below height 30, our numerical lists find 3 ordinary zeros, 11, 21
and 28 quadratic zeros, and 55 positive Artin heights counted twice. That makes
**173 positive field zeros with multiplicity**. There is also a double central
zero. The first 2000 arithmetic coefficients match the factor formula exactly.
The zero picture and sampled winding agree; these new combined checks are not
interval-certified completeness proofs and PARI internally factors the field
function, so it is not an independent computation of an unfamiliar object.

**[known]** The central example shows why bigger zeros do not all flow backward:
our Artin function must vanish at one half, so the field function vanishes there.
Ordinary zeta at one half is about −1.46035, not zero.

**[project]** A later independent interval audit strengthens that finite picture:
Arb enclosed the ordinary and all three quadratic contours and bracketed every
zero. Combined with a fresh rerun of our complete Artin certificate, this gives the
factored count through height 30 rigorously. It is not a new direct field-zeta
calculation; the Artin algorithm is an explicit source dependency. The initial sampled
mesh is still only a sample.

## 2. Tensoring can build a real bridge—but not by tensoring old zeros

**[known]** Tensor our two-dimensional Q8 representation with its dual. The
result splits into the trivial character and the three quadratic characters.
Its L-function therefore really does contain ordinary zeta. Odd tensor powers
just repeat the Artin factor; even powers contain the ordinary-zeta factor.
The symmetric square, by contrast, has only the three nontrivial quadratics.
So “larger representation” is not itself the answer: its exact decomposition is.

**[known]** This does not mean the original Artin zeros determine the tensor
zeros. The tensor operation combines prime-by-prime arithmetic data, not the
old list of zero locations. There are parallel Rankin–Selberg constructions for
automorphic L-functions; in proved adjoint-holomorphy cases they give the same
kind of downward bridge. The general-rank no-cancellation condition must not be
silently assumed.

**[known / project exact check]** There is another pleasant wrinkle: the fourth
symmetric power of our Q8 representation does contain ordinary zeta—twice—even
though its symmetric square does not. We checked the full formula through power
24. That larger function has 66 positive zeros with multiplicity below height
30, from its factors. The exact decomposition, not the word “symmetric,” is what
makes the return ticket valid.

## 3. The quarter-lines really do say something about ordinary RH

**[known]** The quaternion ideal zeta contains zeta at twice the input and zeta
at twice the input minus one. That compresses and shifts ordinary zeros. The
statement “all its nonreal zeros lie on the three lines one quarter, one half
and three quarters” is equivalent to ordinary RH. The middle line comes from
finite local factors; the outside two come from ordinary zeta.

**[known]** This is genuine transfer, but a reformulation rather than new proof:
we would still have to establish the entire three-line claim. Our small finite
checks illustrate the factorization, not its unknown global zero-location claim.

## 4. E8 gives a particularly useful surprise

**[known]** The E8 lattice zeta is an elementary prefactor times ζ(s)ζ(s−3).
Its mirror centre is 2. Yet its **entire centre line is zero-free**! The completed
value there is a negative positive-factor expression. At the centre itself the
completed value is −5/6; the uncompleted lattice sum has continued value −5π²/6.
Those are different normalizations of the same object.

**[known]** Under ordinary RH the nonreal E8 zeros sit at real parts one half
and three and a half, not at the mirror centre. Conversely, proving those two
lines contain every nonreal E8 zero would prove RH by the factors. Thus E8 both
supplies a real return ticket and warns against mistaking a symmetry centre for
a compulsory home for zeros.

## 5. Elsewhere, RH really has been proved

**[known]** Count points on a smooth projective curve over a finite field. Its
zeta function has zeros exactly on a particular circle. This is Weil's theorem
(Hasse for elliptic curves). The proof has extra structure ordinary zeta does
not yet have in the needed form: Frobenius acts as a genuine geometric operator,
and intersection inequalities control it.

**[project]** For y²=x³+x+1 over the five-element field we counted 9 points,
including infinity. Its numerator is 1+3T+5T². Its zeros are
−0.3 ± 0.331662479i, both exactly at distance 1/√5 from the origin. Counting over
larger extension fields checks the same two eigenvalues, not a growing collection
of unrelated zeros.

**[known / project illustration]** Deligne proved the appropriate finite-field
RH in every geometric dimension. The word “appropriate” matters: different
cohomological weights occupy different circles. For our curve times itself—a
surface—we computed two zero circles, of radii 5^(−1/2) and 5^(−3/2). There is
no single universal circle for all dimensions. The cohomological trace formula,
Frobenius, and Deligne's geometric/monodromy methods do real work; just having
more axes does not.

**[known]** On a compact hyperbolic surface, Selberg zeta has a genuine spectral
relationship to a self-adjoint Laplacian. Large-enough eigenvalues put zeros on
the halfway line. Small positive eigenvalues can give real exceptional zeros
off that line, so an unqualified “all Selberg zeros satisfy RH” would be wrong.
Likewise, finite-field RH for local factors does not settle the global
Hasse–Weil L-function of an elliptic curve over the rationals.

**[speculative]** Deninger's cohomological/foliation program and Connes–Consani's
arithmetic/scaling-site program seek an arithmetic replacement for parts of this
geometric machinery. They are distinct programs with real component theorems,
not completed proofs of ordinary RH. A proved positive pairing or a complete
spectral identity would be the missing return ticket.

## 6. Statistics are a valuable clue, not a passport for every zero

**[project]** We computed the first 500 ordinary-zeta zeros and compared their
unfolded pair spacings with the GUE curve from random matrices. The average
adjacent unfolded gap is .999693—close to its chosen unit scale. The histogram
shows the familiar small-gap suppression. The exact method and its limitations
are saved with the data; the larger run, if completed, has its own result file.

**[known]** Montgomery, Rudnick–Sarnak and Katz–Sarnak provide much deeper
restricted theorems, symmetry classifications and conjectures. Bulk spacing is
not the same statistic as low-lying zeros in a family. None of these patterns
alone locates every ordinary zero: finitely many exceptional off-line zeros can
leave a limiting bulk histogram unchanged.

## 7. Three very different infinite spaces give exact return tickets

**[known]** **Li coefficients:** a sequence of real numbers is nonnegative at
every index exactly when RH holds. A finite positive list is not enough. For
our field function, Li coefficients add across the factors; a positive total
at one index does not prove each summand positive.

**[known]** **Weil positivity:** RH is equivalent to positivity of the right
explicit-formula quadratic form for every admissible test function. A promising
higher-algebra construction would have to recover precisely that form and prove
its positivity. A generic squared norm is positive but is not automatically the
same form.

**[known / project]** Li and Weil even meet in an exact matrix identity. We
interval-certified a 32-dimensional compression as positive using 50 rigorously
computed Li coefficients. Its tiny final pivot is about 10^−110; an ordinary
floating-point calculation invents a negative pivot through rounding. This is
both a concrete finite positivity result and a warning: thirty-two directions
are not all the directions in the infinite space.

**[known]** **Nyman–Beurling:** RH is equivalent to a particular approximation
problem in a Hilbert space. Make the error tend rigorously to zero and information
returns to every zero. We wrote out a quantitative obstruction: an off-line zero
β+it with β>1/2 prevents the squared error in a tail-cancelled subspace from
falling below (2β−1)/|β+it|². One finite error bound becomes weaker at large height;
a convergent sequence would rule out every fixed off-line zero.

**[known]** **Heat flow:** the de Bruijn–Newman family moves zeta into an extra
parameter. A constant Λ marks where every zero becomes real. Ordinary RH is
now equivalent to Λ=0: Rodgers–Tao proved the ≥0 half, and ≤0 is the remaining
RH statement. Seeing all-real zeros at a positive time does not automatically
let us run the result backwards to zero.

**[known]** **Jensen polynomials:** all the relevant polynomials having real
roots is another exact RH criterion. Fixed-degree eventual real-rootedness is
a proved and beautiful theorem, but it is not uniform control of every degree
and shift. The quantifiers are the bridge's load-bearing beams.

**[project]** We made the approximation bridge concrete with interval arithmetic:
using integer generators through 30, our chosen approximant has squared error
.014556733577. For the first generator alone it is exactly 1−log 2. These finite
bounds are much weaker than known zeta results; the value is seeing an exact
return mechanism operate, not claiming a better zero-free region. The missing
step remains a rigorous sequence whose error tends to zero.

## What tonight's six differences give us

**[project]** Direction-sensitive zeros, forced Artin centres, bracket-sensitive
star products, and holes at infinity reveal genuinely different structures.
None currently supplies an independent restriction on ordinary-zeta real parts.
Quarter-lines and E8 do transfer through their factors. An intrinsic lift into
quaternions or octonions transfers through exact restriction—but it restates
rather than proves the old zero-location problem. This is a ledger of what we
can presently infer, not a declaration that the unexplored routes are exhausted.

**[speculative]** The next serious search is for something position-sensitive:
a form, operator, approximation theorem or factor identity whose ordinary slice
is exactly the known zeta object. A beautiful analogy can tell us where to look.
The return ticket tells us when we have actually arrived back at the number line.

## The longer statistical run finished

[project] We also computed the first **5000** ordinary zeros, up to height
5447.861998. There are 17492 close pairs in the chosen unfolded window. The
plotted discrepancy from the random-matrix curve falls from .09349 for 500
zeros to .03809 for 5000. That is a descriptive comparison, not an RH test.
A finite exceptional zero set can be invisible even to a perfect limiting
statistical law.

[project] The arithmetic audit was extended to all 5000 stored Artin coefficients
and the field-factor convolution through 5000. The Artin contour was also rerun completely in 221.824 seconds. Every non-timing
output matches the frozen earlier certificate. It is a reproduced proof, not
an independent numerical algorithm.

## A classical example where a larger family really did help

[known] There is a simple positive relation between ζ at three shifted points:
ζ(σ)^3 |ζ(σ+it)|^4 |ζ(σ+2it)|≥1 for σ>1. It comes from the exact identity
3+4cos θ+cos 2θ=2(1+cos θ)^2. If there were a zero at 1+it, the left side would
tend to zero as σ approached 1, contradicting that lower bound. Thus the whole
line Re(s)=1 is zero-free. This is established mathematics, not a project result.

[known] Why does it stop there? The positive expansion is justified to the right
of 1. It does not automatically survive inside the critical strip. This is a
particularly clear example of the bridge we want: multiple points, an exact
identity tying them together, and an inequality strong enough to exclude zeros.
See S23 and the complete short proof in the ledger.

## Back on the actual number line

[known] There is one last exact bridge. If the larger Galois-field function
satisfied RH, ordinary ζ would satisfy RH too. That would imply prime-power
counting error at most a constant times √x log²x, and prime-counting error at most
a constant times √x log x, uniformly along the positive number line. Von Koch's
1901 original paper gives the prime-counting implication; the ledger spells out
the converse analytic bridge for prime powers.

[project] Our finite-height certificates do not establish that all-height premise.
To get a useful finite-range prime estimate from a finite zero list, one must
also bound the omitted explicit-formula tail. That is a concrete next task,
not a consequence we silently attach to the pictures.
