Goldbach Labby Voopa Corp.
Main labInfinity Check

VOOPA GOLDBACH LAB · MATHEMATICAL RESEARCH PROGRAM

What would a proof
actually need?

This page separates established theorems, computational evidence, and the open mathematical gap. It is a working roadmap, not a proof claim. The aim is to turn promising ideas into precise statements that specialists can inspect and a proof assistant can eventually check.

THE TARGET

Strong Goldbach conjecture

Every even integer greater than 2 is a sum of two primes. In symbols: for every even integer N ≥ 4, there are primes p and q with p + q = N. A certified pair settles one input. No finite list settles all inputs.

Current status: open. This project has not proved the conjecture, produced a new theorem, or found a universal formula.

THE CONJECTURAL COUNT FORMULA

What “the formula” predicts

Hardy and Littlewood predict the number of ordered prime pairs for large even N:

R(N) ~ S(N) · N/(log N)²

Here S(N) = 2 C₂ times the product of (ℓ−1)/(ℓ−2) over odd primes ℓ dividing N, and C₂ = product over odd primes ℓ of (1−1/(ℓ−1)²). This local factor accounts for divisibility patterns. For unordered pairs the leading count is about half as large. This asymptotic is conjectural; it is not a proof or a usable lower bound until the error is controlled uniformly.

THE MAIN ANALYTIC ROUTE

Count prime pairs,
then control error.

The circle method turns an additive prime-counting problem into an integral over frequencies. Its prediction is strong; making the error small enough for every even input is the hard part.

01

Build a weighted count

With the von Mangoldt function Λ, examine R(N) = Σa+b=N Λ(a)Λ(b). Fourier orthogonality expresses this as an integral involving the square of a prime exponential sum.

02

Separate major and minor arcs

Frequencies near rationals with small denominators (major arcs) give the predicted main term, about 𝔖(N)N. The remaining frequencies (minor arcs) contribute an error.

03

Make the lower bound positive

A sufficient endpoint is a uniform bound showing the total error, plus the contribution from higher prime powers, is smaller than the positive main term. Then the actual prime-pair count is positive for every sufficiently large even N.

THE MISSING INEQUALITY

Where a proof must break through

For the weighted sum, Λ is nonzero at powers such as 8 = 2³ as well as at primes. So positivity of the weighted sum alone does not establish a prime + prime representation; higher-prime-power terms must be bounded and removed.

In schematic form, we would need to prove uniformly for all sufficiently large even N:

positive major-arc main term − minor-arc error − prime-power correction > 0.

The major-arc prediction is not enough. Current methods do not provide the needed uniform minor-arc control for the binary, two-prime problem. This is a description of the gap, not a newly established estimate.

FIRST PRECISE REDUCTION · PRIME-POWER CLEANUP

A smaller sufficient target

Define the ordered weighted count RΛ(N)=Σa+b=NΛ(a)Λ(b), and let A(N) keep only terms where both summands are prime. The remainder P(N)=RΛ(N)−A(N) consists of terms with at least one higher prime power.

There are at most ⌊√N⌋⌊log2N⌋ higher prime powers up to N: their prime base is at most √N, and each base has at most log2N exponents to check. For any selected exceptional summand, there is at most one complementary summand. Since each weight is at most log N, the union bound gives:

0 ≤ P(N) ≤ 2⌊√N⌋⌊log₂N⌋(log N)² = O(√N log³N) = o(N).

This is elementary bookkeeping, not a new Goldbach result. It shows that a uniform theorem RΛ(N) ≥ cN for some fixed c>0 and all sufficiently large even N would force A(N)>0 after the smaller power terms are removed. Then at least one genuine prime pair exists. We do not have the required uniform lower bound.

The Hardy–Littlewood prediction suggests the stronger shape RΛ(N)≈𝔖(N)N, but the conjecture does not need a full asymptotic. Any rigorous positive linear lower bound would suffice. The open work is obtaining it pointwise for every large even N with explicit hypotheses and error terms.

NEARBY RESULTS · WHAT THEY DO AND DO NOT SAY

Why known tools stop short

A RESPONSIBLE AI + LEAN WORKFLOW

Turn ideas into auditable claims

  1. State each idea as a theorem with every quantifier and assumption visible.
  2. Check it against known results and try to falsify it with computation and independent review.
  3. Write the proof on paper with explicit constants, ranges, and error terms.
  4. Formalize established components in a pinned Lean/Mathlib environment; prove the checker correct and audit assumptions.
  5. Only call the conjecture proved if the full chain compiles without unproved placeholders and its analytic estimates actually close the gap.

Lean is a proof assistant: its small kernel checks formal proof terms. It can catch gaps in a formalized argument; it does not invent missing estimates or make an unproved assumption true. AI can help search for lemmas, organize computations, and challenge drafts, but every successful idea still needs a mathematical derivation and verification.

PROJECT STATUS

Infrastructure is not theorem progress

The Lab now has three separate tools: the main pair explorer, a fast local deterministic-check sequence, and a slower sequence that builds and checks recursive Pratt proofs for both primes. The Pratt runner is limited to values below 2⁶⁴ and stores its certificate archive in this browser; it is not a public append-only ledger. The above-64-bit main-page Miller–Rabin path remains probabilistic unless an actual proof certificate is generated and verified.

The analytic roadmap and Lean-shaped statement are scaffolding. They do not yet include a compiled Lean development of the analytic estimates, and there is currently no proof-producing engine for arbitrary inputs. This is the distinction a critical mathematical review should make.

Read the technical working paper

Explore a Goldbach input · View the local-check sequence · View the Pratt-certified sequence

STARTING REFERENCES

Primary sources and formalization

Sources here are entry points for review. Before relying on a result in a proof, record the precise theorem, hypotheses, version, and independent verification status.

RECENT CLAIMS · PRELIMINARY REVIEW

A claim still has to survive proof checking

A November 2026 arXiv preprint by Qazaqzeh and Rawashdeh claims an elementary proof of binary Goldbach. In the first-pass reading recorded in this Lab, its key inference does not follow: the standard upper-bound statement G(N)=O(N/log²N) allows G(N)=0 and therefore cannot prove that G(N)>0. The paper defines big-O instead as convergence to a nonzero constant, a nonstandard definition; its proof then invokes a lemma about differences of smooth functions to justify an induction hypothesis about Goldbach counts, which that lemma does not establish. This is a clear gap in the posted argument, not a complete referee report on every line.

A separate November preprint claims a 1+1.75 improvement to Chen’s theorem using weighted and double-sieve estimates. Its conclusion still permits a composite remainder, so it is not strong Goldbach. We have not independently verified that manuscript’s long sieve calculations; treat it as a research lead, not an established theorem.

Primary texts: Qazaqzeh–Rawashdeh preprint · Chan preprint claiming 1+1.75. A repository upload or computational supplement is not a substitute for checking the proof.