On the the primes between a quadratic pair

The first “the” is the persona. The second is the shadow. Between them lies an error term.

I began with two squares:

\[ Q_n^- = n^2, \qquad Q_n^+ = (n+1)^2. \]

Between them I wanted a prime.

Not a prime for almost every \(n\), not a prime with high probability, and not a prime after averaging away the embarrassing cases. I wanted the stubborn statement

\[ \mathsf{QP}(n): \qquad \vartheta\!\left((n+1)^2\right)-\vartheta(n^2)>0, \]

where

\[ \vartheta(x)=\sum_{p\le x}\log p. \]

This is the smallest door through which I entered the largest room. The distance between the squares is only

\[ (n+1)^2-n^2=2n+1, \]

yet the question summons the entire distribution of the primes. I thought I was looking into a narrow interval. The interval looked back.

That was my first encounter with the mathematical unconscious.

Who I am

I am not, at present, the person who has proved the Riemann Hypothesis. That would be a particularly important detail to leave out of the footnotes.

I am the person who has tried to construct the conditions under which a proposed proof could not conceal its own mistakes. I have worked through grwtsk/math as a mathematical interface and through Noeaaeue as an architecture for decomposition, delegation, verification, and recomposition. My work has therefore been partly mathematical, partly computational, partly formal, and partly an extended argument with my own enthusiasm.

Existentially, I am an interface between conjecture and audit.

I stand between the sentence

\[ \text{“These structures appear to be the same”} \]

and the demand

\[ \text{“Provide the map, its domain, its codomain, its hypotheses, and its proof obligations.”} \]

I have become a cartographer of missing morphisms.

The primes supplied the orientation. A prime cannot be factored without becoming something else. In Jungian terms, the prime has individuated. It has rejected decomposition and become its own whole number. I found this admirable, although mathematically inconvenient.

My own individuation has been less efficient. Every time I resolved one complex, another complex number appeared.

The critical line,

\[ \operatorname{Re}(s)=\frac12, \]

became the axis around which this work turned. The nontrivial zeros

\[ \rho=\beta+i\gamma, \qquad 0<\beta<1, \]

possess the familiar symmetries

\[ \rho\longmapsto \overline{\rho}, \qquad \rho\longmapsto 1-\rho. \]

The Riemann Hypothesis says that all of these zeros have completed their analytic individuation:

\[ \mathrm{RH} \iff \forall \rho,\quad \beta=\frac12. \]

Every zero wants to be centered. This is either profound or merely good typography.

What I have tried

I first tried geometry.

I drew circles, ellipses, chords, antipodal points, tangent returns, radial centers, and sweeping rotations. I asked whether a quadratic pair could be made to pass through a family of ellipses and whether the coincidence of split points at the circular case could expose the symmetry of the critical line.

The geometric instinct was not arbitrary. A quadratic equation naturally produces paired roots, involutions, discriminants, and branch behavior. A circle offers a global symmetry; an ellipse offers a deformed one. The passage

\[ \text{quadratic} \longrightarrow \text{quartic} \longrightarrow \text{local pieces} \longrightarrow \text{sheaf-theoretic gluing} \]

seemed capable of turning a visual coincidence into a structural invariant.

I then tried arithmetic recursion. I worked from the two faces of the zeta function:

\[ \zeta(s)=\prod_p\frac{1}{1-p^{-s}}, \qquad \frac{1}{\zeta(s)} = \sum_{n=1}^{\infty}\frac{\mu(n)}{n^s}. \]

The Euler product presents the primes as multiplicative atoms. Möbius inversion presents cancellation as an arithmetic reversal. I asked whether the movement between these descriptions could be made recursively self-similar—something fractal, Mandelbrot-like, and centered around \(1/2\).

This was psychologically irresistible. The zeta function had a persona, an Euler product, and a shadow, its zeros. Möbius inversion was the analyst in the room, repeatedly asking the integers how their factorization made them feel.

I tried graph theory. I considered graph sandwiches,

\[ G_{\mathrm{lower}} \subseteq G \subseteq G_{\mathrm{upper}}, \]

spindles, colorings, spectral bounds, cellular incidence, and combinatorial types. The hope was that arithmetic admissibility might be represented as a coloring or extension problem, with forbidden configurations corresponding to violations of the desired analytic symmetry.

I tried Hodge-theoretic language. Local data, boundary operators, harmonic representatives, and cohomological obstructions offered a model for separating genuine global structure from local noise. I asked whether the flow of zero configurations could be approximated by a Hodge decomposition:

\[ C^k = \operatorname{im} d \oplus \ker \Delta \oplus \operatorname{im} d^\ast. \]

Perhaps the unwanted displacement from the critical line could be interpreted as an exact or coexact defect, leaving the harmonic component centered.

I tried \(K\)-theory as a generalized adapter. If two mathematical realizations carried different local representations but the same stable class, then perhaps a framed port could translate one proof obligation into another:

\[ K_\ast(A)\longrightarrow K_\ast(B). \]

This joined the work on complete type theories, framed interfaces, fiber-port mechanisms, and operator lifting. A theorem could become a typed object. A proof could become a transport law. An obstruction could become a failure to inhabit a type.

I tried four-phase projective embeddings. This appealed to the Jungian quaternity: four phases, four directions, four corners of the mandala. A quaternity always looks suspiciously like a theorem when it is drawn inside a circle. Unfortunately, the universe does not award mathematical credit for archetypal completeness.

I tried random matrix theory and the Gaussian unitary ensemble. The local statistics of high zeta zeros resemble eigenvalue statistics from GUE. I connected this to scattering, spectral heights, noncommutative geometry, and the possibility of an operator whose spectrum records the ordinates \(\gamma\).

I tried to relate this to Connes-style trace formulas and to heat-flow arguments. I asked whether a Hodge approximation of a spectral flow could produce a monotone quantity that moved a configuration toward the critical line without losing the arithmetic trace.

Finally, I tried to make the entire search formally accountable. I used proof capsules, source-bound claims, exact identities, hostile fixtures, module-specific outcomes, Lean declarations, independent reviews, and swarm roles. I moved from the vague sentence “the tests passed” toward a relation such as

\[ \mathsf{ModulesSucceeded}(M,R) \iff \forall m\in M,\; \exists! r\in R \]

such that

\[ \operatorname{sourceID}(r)=\operatorname{sourceID}(m) \quad\land\quad \operatorname{status}(r)=\mathrm{PASS}. \]

One unique pass for each required identity: not a nearby pass, not a root-level pass, not two conflicting passes, and not a pass produced by an unrelated artifact wearing the correct hat.

The collective unconscious, I discovered, is a terrible place to store dependencies.

What has worked, and why

The architecture has worked.

It has not worked in the sense that it has produced a proof of the Riemann Hypothesis. It has worked in the more immediate and necessary sense that it has exposed where a proof would otherwise be pretending.

The geometric work has been useful when it identifies an invariant rather than merely illustrating a resemblance. A visualization can reveal that two constructions share an involution, a conserved orientation, a singular case, or a failure of continuity. It becomes mathematically useful when the picture induces a precise statement such as

\[ F\circ \iota=\iota\circ F \]

\[ \operatorname{Fix}(\iota) = \left\{s:\operatorname{Re}(s)=\frac12\right\}. \]

The sheaf language has worked as an audit of local-to-global reasoning. It forces me to state what the local sections are, what overlaps mean, and whether compatible local data actually glue. Many beautiful ideas fail at the final condition:

\[ s_i|_{U_i\cap U_j} = s_j|_{U_i\cap U_j}. \]

They agree in spirit but not on the overlap. Mathematics is rather strict about shared custody.

The \(K\)-theoretic and type-theoretic work has worked as an organizational discipline. It distinguishes a representation from the invariant it is meant to preserve. It also exposes when I have called something an “adapter” merely because I have not yet defined the translation.

The graph work has worked when it converts analytic ambiguity into a finite obstruction problem. A graph can make incompatibility visible. A spectral graph approximation can reveal which estimates are stable under discretization. It can also provide adversarial examples against an overconfident conjecture.

The formal verification work has worked best of all, because it has made vacuity visible. An always-rejecting checker may be sound:

\[ \mathsf{Accept}(x)\Rightarrow P(x), \]

simply because \(\mathsf{Accept}(x)\) is never true. This is the logical equivalent of a nightclub with perfect security because it has no door.

A useful verification boundary therefore needs both soundness and non-vacuity:

\[ \exists x,\quad \mathsf{Accept}(x), \]

together with exact source identity, complete coverage, and independent reproduction.

The most important thing that has worked is the conversion

\[ \text{analogy} \longrightarrow \text{typed conjecture} \longrightarrow \text{explicit obligation} \longrightarrow \text{falsifiable test}. \]

This does not prove the conjecture. It prevents the metaphor from receiving tenure.

What has not worked, and why

The grand synthesis has not yet worked.

The circle, the ellipse, the graph sandwich, the Hodge decomposition, the \(K\)-class, the scattering height, the GUE spectrum, and the four-phase embedding may all illuminate the same problem. Illumination is not identification.

Formally,

\[ A\sim B \not\Rightarrow A=B, \]

and even

\[ \operatorname{Stat}(A)=\operatorname{Stat}(B) \]

does not provide a canonical map between them.

The GUE correspondence offers powerful statistical evidence about zero spacings. It does not by itself imply that every zero satisfies

\[ \operatorname{Re}(\rho)=\frac12. \]

Average spectral behavior is not a pointwise theorem. A crowd can have excellent statistics while one individual zero behaves badly. The unconscious is collective; the counterexample only needs to be personal.

The operator approach has not worked because the decisive property is too easily placed inside the assumptions. Suppose I seek a self-adjoint operator \(H\) with

\[ \operatorname{Spec}(H)=\{\gamma:\zeta(\tfrac12+i\gamma)=0\}. \]

Because a self-adjoint operator has real spectrum, this would explain why the zeros lie on the critical line. But constructing precisely that operator, with the required trace formula and domain theory, is not an administrative detail. It is the dragon.

If I assume the dragon is already house-trained, I have not completed the quest. I have changed its type annotation.

The Connes-style route has not yet worked because a trace formula must do more than reproduce the explicit formula symbolically. It must provide a positivity mechanism strong enough to constrain the zeros. The arithmetic and spectral sides must agree on an appropriate test-function space, with all regularizations, boundary terms, and domains controlled.

The Hodge and heat-flow routes have not worked because an analogy to decomposition is not yet a monotonicity theorem. A successful flow needs a quantity \(E(t)\) satisfying something like

\[ \frac{d}{dt}E(t)\le 0, \]

together with a characterization of equality and a proof that the limiting configuration corresponds to the desired zero geometry. Without those facts, “flowing toward the critical line” is simply a very polite way to move symbols across a page.

The fractal recursion has not worked because self-similarity can preserve an error as efficiently as it preserves a truth. A recursive system is not automatically a convergent system. One needs a metric, a contraction, a fixed-point theorem, and a proof that the fixed point is relevant:

\[ d(Tx,Ty)\le c\,d(x,y), \qquad 0\le c<1. \]

Otherwise the construction may be fractally wrong at every scale, which is at least consistent.

The formal work has also taught me a less glamorous lesson: source-level declarations that have not been accepted by the intended kernel remain source-level declarations. A packet may be careful, bounded, and internally coherent while still being

\[ \mathrm{NOT\_RUN} \neq \mathrm{KERNEL\_VERIFIED}. \]

A proof assistant does not accept promissory notes, even beautifully formatted ones.

What I am trying now

I am returning to the quadratic pair, but with the entire architecture concentrated on one exact expression.

Let

\[ \psi(x)=\sum_{p^k\le x}\log p \]

be the second Chebyshev function. The explicit formula gives, for suitable \(x\),

\[ \psi(x) = x - \sum_{\rho}\frac{x^\rho}{\rho} - \log(2\pi) - \frac12\log(1-x^{-2}). \]

Therefore

\[ \begin{aligned} \Delta_n\psi &= \psi\!\left((n+1)^2\right)-\psi(n^2)\\[3pt] &= 2n+1 - \sum_{\rho} \frac{(n+1)^{2\rho}-n^{2\rho}}{\rho} + R_n, \end{aligned} \]

where the elementary remainder is

\[ R_n = -\frac12 \log \left( \frac{1-(n+1)^{-4}}{1-n^{-4}} \right). \]

This is the center of the present program. The question between the squares has become a question about whether the collective contribution of the zeros can overwhelm the main term \(2n+1\).

To return from prime powers to primes, I must also account for

\[ \Delta_n\vartheta = \Delta_n\psi - \sum_{k\ge 2} \;\sum_{n^2<p^k\le (n+1)^2} \log p. \]

This correction matters. A positive jump in \(\psi\) may be caused by a prime power rather than a new prime. Even the primes have impersonators.

Under RH, writing

\[ \rho=\frac12+i\gamma \]

turns the spectral contribution into an oscillatory expression of the form

\[ \sum_{\gamma} \frac{ (n+1)^{1+2i\gamma} - n^{1+2i\gamma} }{ \frac12+i\gamma }. \]

I am trying to understand its cancellation not merely term by term, but structurally. The random-matrix and scattering pictures suggest that the phases should repel, decorrelate, or cancel. The noncommutative-geometric picture suggests that this sum should arise as a trace. The Hodge picture suggests that the dangerous component might be separated from a harmonic core. The graph picture suggests finite approximants and adversarial boundary cases.

Noeaaeue supplies the decomposition:

\[ \text{explicit formula} \rightarrow \text{spectral model} \rightarrow \text{cancellation estimate} \rightarrow \text{prime-power correction} \rightarrow \text{quadratic interval}. \]

Each arrow must become its own claim, certificate, and checker. The swarm can distribute the work, but it cannot distribute away the missing theorem.

Why what I am trying now will not work

In its current grand form, it will not work because it asks one architecture to solve several logically distinct problems at once.

The spectral operator may reproduce the zeros without proving the required positivity. The GUE statistics may explain average cancellation without giving a uniform bound for every quadratic interval. The Hodge model may control finite cellular approximations without controlling the limiting analytic object in the topology that matters. \(K\)-theory may preserve stable information while discarding precisely the fine spectral location I need.

A universal adapter can become a universal hiding place.

There is also a quantitative obstruction. The standard consequence of RH is an estimate of the general shape

\[ \psi(x) = x + O\!\left(x^{1/2}\log^2 x\right). \]

At \(x=n^2\), this gives an error on the scale

\[ O(n\log^2 n), \]

while the main term between consecutive squares is only

\[ 2n+1. \]

The standard RH error estimate is therefore too large to guarantee positivity in every interval \((n^2,(n+1)^2)\). Even if the Riemann Hypothesis were handed to me in a sealed envelope, the most direct route from it to the prime between every pair of consecutive squares would still be blocked.

This is an important correction to my own narrative. The quadratic-pair problem led me to RH, but it is not merely RH wearing a smaller coat.

My present synthesis also risks circularity. If I define an operator whose self-adjointness is equivalent to all zeros having real spectral parameters, then prove RH by invoking the operator’s self-adjointness, I have factored the conclusion into the premise. That is not prime reasoning.

The four-phase structure may still be valuable, but Jungian completeness is not mathematical completeness. A mandala can organize the psyche. It cannot, without further argument, estimate a zero sum.

My shadow keeps changing sign. That is poetically appropriate and analytically unhelpful.

What I will do next

I will stop asking the whole architecture to prove everything simultaneously.

The next step is to separate the logical ladder:

\[ \mathsf{EF} \longrightarrow \mathsf{SC} \longrightarrow \mathsf{PC} \longrightarrow \mathsf{QP}, \]

where:

\[ \mathsf{EF} = \text{the explicit-formula identity}, \]

\[ \mathsf{SC} = \text{a precisely quantified spectral-cancellation estimate}, \]

\[ \mathsf{PC} = \text{control of prime-power contributions}, \]

and

\[ \mathsf{QP} = \vartheta((n+1)^2)-\vartheta(n^2)>0. \]

I will prove the arrows that can be proved without pretending that \(\mathsf{SC}\) follows automatically from RH. I will state exactly how much cancellation is required, in what norm, over what range, and with what smoothing.

A first tractable target may be averaged rather than pointwise:

\[ \sum_{n\le N} \left| \Delta_n\psi-(2n+1) \right|^2 \le \mathcal E(N), \]

with an explicit \(\mathcal E(N)\). From there I can estimate the exceptional set

\[ \left\{ n\le N: \Delta_n\vartheta\le 0 \right\}. \]

This will not prove the desired statement for every \(n\), but it will identify the exact distance between statistical control and pointwise control. That distance is where the real theorem must live.

I will also replace the universal operator with a sequence of bounded questions. Which test functions produce a rigorous prime–zero trace identity? Which positivity statements are genuinely stronger than RH? Which graph approximations preserve the relevant quadratic form? Which Hodge decomposition survives the analytic limit? Which \(K\)-theoretic data are informative, and which are too stable to remember where a zero actually sits?

Each claim will have a non-vacuous witness and an independently reproducible check. Formalization will follow the mathematics rather than standing in for it. A source packet will remain marked as source-only until a kernel accepts it. A numerical experiment will remain evidence about a range, not an induction principle over infinity.

The next object I want is not a cathedral. It is a hinge.

I began by asking whether a prime lived between two squares. I thought the interval was empty space. It turned out to contain geometry, spectral theory, graph theory, Hodge theory, \(K\)-theory, type theory, noncommutative geometry, random matrices, formal verification, and several versions of myself.

The quadratic pair became my pair of opposites: the local and the global, the visual and the formal, the conjectural and the verified, the persona and the shadow. The critical line became not merely a location for zeros but a demand for balance.

I have not yet solved the Riemann Hypothesis. I have learned something more immediately useful: where my attempted proof would have to stop lying.

My next move is therefore to factor myself out of the argument.

The primes, after all, have already shown me how.