# A proof target for binary Goldbach: the weighted lower-bound route

**Working note · 22 November 2026**  
**Status:** a conditional reduction and a research plan. It contains no proof of strong Goldbach and no new analytic estimate.

## Abstract

The Hardy–Littlewood heuristic predicts that an even integer (N) has about \(\mathfrak S(N)N/(\log N)^2\) ordered representations as a sum of two primes. A proof does not require the full asymptotic. This note isolates a weaker sufficient target: a uniform positive linear lower bound for the von Mangoldt convolution

\[
R_\Lambda(N)=\sum_{a+b=N}\Lambda(a)\Lambda(b).
\]

The convolution also counts prime powers. We give an elementary upper bound for those unwanted terms of order \(\sqrt N(\log N)^3\), hence (o(N)). Therefore any unconditional estimate \(R_\Lambda(N)\ge cN\), with fixed (c>0) for every sufficiently large even (N), implies a prime-plus-prime representation for every such (N). The unsolved step is the pointwise analytic lower bound; the reduction is bookkeeping, not progress that closes that gap.

## 1. Exact target and conventions

For an even integer (N\ge4), define the ordered unweighted count

\[
G(N)=\#\{(p,q): p,q\text{ prime},\ p+q=N\}.
\]

The strong Goldbach conjecture is \(G(N)>0\) for every even (N\ge4\). Ordered pairs count ((p,q)) and ((q,p)) separately when (p\ne q).

Let \(\Lambda(m)=\log p\) if (m=p^k) for a prime (p) and integer (k\ge1), and let \(\Lambda(m)=0\) otherwise. Define

\[
R_\Lambda(N)=\sum_{\substack{a,b\ge1\\a+b=N}}\Lambda(a)\Lambda(b),\qquad
A(N)=\sum_{\substack{p,q\text{ prime}\\p+q=N}}\log p\log q.
\]

Write (H(N)) for the set of integers (m\le N) that are prime powers (p^k) with (k\ge2). Set

\[
P(N)=\sum_{\substack{a+b=N\\a\in H(N)\text{ or }b\in H(N)}}\Lambda(a)\Lambda(b).
\]

The cases in (P(N)) are counted once, including pairs for which both entries are higher prime powers. Since the summands are nonnegative,

\[
R_\Lambda(N)=A(N)+P(N).
\]

## 2. Prime-power correction lemma

**Lemma.** For (N\ge4),

\[
0\le P(N)\le 2\lfloor\sqrt N\rfloor\lfloor\log_2N\rfloor(\log N)^2
=O(\sqrt N(\log N)^3)=o(N).
\]

**Proof.** If (p^k\le N) and (k\ge2), then (p\le\sqrt N). There are at most \(\lfloor\sqrt N\rfloor\) possible prime bases, because this also bounds the number of all positive integer bases. For each base, (k\le\log_2N). Unique factorization makes each such prime power correspond to one pair ((p,k)), so

\[
|H(N)|\le\lfloor\sqrt N\rfloor\lfloor\log_2N\rfloor.
\]

For each selected exceptional coordinate (a\in H(N)), the equation (a+b=N) determines at most one (b); the same count applies when the exceptional coordinate is (b). The union bound therefore gives at most (2|H(N)|) ordered pairs in (P(N)). For (1\le m\le N), (0\le\Lambda(m)\le\log N), so each product weight is at most \((\log N)^2\). Combining these bounds proves the displayed inequality. Finally, \((\log N)^3/\sqrt N\to0\), proving the (o(N)) statement. \(\square\)

The bound is intentionally crude. Its role is only to make the logical endpoint explicit; improving it does not address the principal difficulty.

## 3. Conditional reduction to a lower bound

**Proposition.** Suppose there are constants (c>0) and (N_0) such that every even (N\ge N_0) satisfies

\[
R_\Lambda(N)\ge cN.
\]

Then strong Goldbach holds for every even (N\ge N_1), for some (N_1\ge N_0).

**Proof.** By the lemma, choose (N_1) so that (P(N)<cN) for every (N\ge N_1). Then

\[
A(N)=R_\Lambda(N)-P(N)>0.
\]

Every term in (A(N)) is positive, so (A(N)>0) means at least one prime pair sums to (N). \(\square\)

The remaining finite interval (4\le N<N_1) could in principle be checked with proof-producing certificates. The proposition is a conditional implication, not evidence that its hypothesis is known.

### Refining the target: the linear bound is not necessary

The hypothesis `R_Λ(N) ≥ cN` is a clean, Hardy–Littlewood-shaped sufficient condition, but it is stronger than Goldbach asks for and should not be called the weakest target. The exact endpoint from the decomposition is simply `R_Λ(N) > P(N)`, equivalently `A(N) > 0`.

Using only the explicit crude bound in the lemma, another sufficient analytic statement is

```text
R_Λ(N) > 2⌊√N⌋⌊log₂N⌋(log N)²
```

for every sufficiently large even `N`. This threshold is `o(N)`, so it is strictly weaker than demanding a fixed positive proportion of `N`. It remains a very difficult pointwise lower bound for a binary prime correlation, and we have no proof of it. A refined estimate for `P(N)` could lower the sufficient threshold further. The hierarchy matters: `R_Λ(N) ≥ cN` is a convenient strong target; it is neither known nor logically required by Goldbach.

## 4. Relation to Hardy–Littlewood and the missing estimate

For even (N), Hardy–Littlewood predict

\[
G(N)\sim\mathfrak S(N)\frac{N}{(\log N)^2},\qquad
\mathfrak S(N)=2C_2\prod_{\substack{\ell\mid N\\\ell>2\\\ell\text{ prime}}}\frac{\ell-1}{\ell-2},
\]

where (C_2=\prod_{\ell>2,\,\ell\text{ prime}}(1-(\ell-1)^{-2})). In the weighted formulation, the predicted main term is of order \(\mathfrak S(N)N\). Because each factor in the finite product exceeds one and (C_2>0), this heuristic suggests a positive linear lower bound for (R_\Lambda(N)). It does not prove one.

The analytic target can be stated without demanding the full Hardy–Littlewood asymptotic:

\[
\exists c>0, N_0\ \forall N\ge N_0:\quad 2\mid N\Longrightarrow R_\Lambda(N)\ge cN.
\]

A circle-method proof would need a major-arc contribution with a uniform positive linear lower bound, and a minor-arc plus truncation error smaller than that lower bound for each even (N). An estimate that holds only on average, outside an exceptional set, or under an unproved hypothesis does not establish this statement. The prime-power lemma above handles the conversion from weighted mass to an actual prime pair once the lower bound exists.

## 5. Research tasks and stop conditions

1. **Audit the exact binary major-arc formulas.** Fix the weight and smoothing convention, then record the singular series, truncation, and every uniformity condition.
2. **Quantify the binary minor-arc obstruction.** For the square of the prime exponential sum, determine the best proved pointwise bound on the full minor-arc integral and compare its size with the required (cN). Preserve the complex phase; replacing it with an absolute-value estimate may erase the cancellation needed.
3. **Check adjacent inputs, not just averages.** Any proposed recurrence for (G(N)), weighted counts, or sieve coefficients must be proved from its exact definitions. Positivity at (N-2) does not imply a lower bound of order (N/\log^2N) at (N).
4. **Keep conditional routes visibly conditional.** RH, stronger distribution in progressions, or a conjectural correlation estimate can be used to derive a conditional theorem, but the assumptions must be isolated and cannot be described as a proof of unconditional Goldbach.
5. **Formalize only proved reductions.** Lean can encode the definitions and check the prime-power cleanup implication after the elementary counting bound is formalized. It cannot supply the missing analytic estimate. Pin the Lean/Mathlib version and audit axioms before claiming a formal result.
6. **Require a novelty gate.** For every proposed lemma, record its exact statement, source search, proof sketch, counterexample tests, independent mathematical review, and formalization status. Numerical tests can refute a candidate claim; they cannot certify its universal quantifiers.

## 6. Present conclusion

The most economical theorem target we have identified is not “prove the Hardy–Littlewood formula exactly,” but “prove a uniform positive linear lower bound for (R_\Lambda(N)).” The higher-prime-power cleanup is elementary and asymptotically negligible. The hard, unresolved part remains the two-prime correlation estimate that would force this lower bound for every sufficiently large even input. No part of this note claims to provide that estimate.

## Starting references

- H. A. Helfgott, [*Major arcs for Goldbach's problem*](https://arxiv.org/abs/1305.2897).
- H. A. Helfgott, [*Minor arcs for Goldbach's problem*](https://arxiv.org/abs/1205.5252).
- H. A. Helfgott, [*The ternary Goldbach conjecture is true*](https://arxiv.org/abs/1312.7748).
- O. Ramaré, [*On Šnirel'man's constant*](https://arxiv.org/abs/2005.12953), for additive prime methods and explicit analytic context.
- Qazaqzeh and Rawashdeh, [*An Elementary Proof of the Binary Goldbach Conjecture*](https://arxiv.org/abs/2611.09007), a recent preprint claim; its claimed argument is not treated here as accepted proof.
