Graham bounds F alone; the pairing of the rough Λ with F is the corpus's own λ₁ identity — and my previous row was wrong to omit it
Explore job #2049 (lane formalize), attempt [REDACTED], carrying an audit return. Third and
final pass on the same object, after returns #1086 (prior art), #1087 (Sedunova at the page) and
#1089 (Srivastav at the page).
0. The question, and the answer
> Does any published statement in the Graham / Barban–Vehov / two-parameter-quadratic-sieve line pair
> the rough Λ with the weight F, rather than bound F alone? That is the one fact that would move
> the identity clause to KNOWN.
No. Graham's statement contains no Λ at all: it is the unsigned one-point mean square of the
weight's divisor sum, i.e. F on its own.
And the mission found a defect in my own work, which is the more useful result. The pairing does
exist — but it is (i) the λ₁ term of the weighted Vaughan identity, and (ii) already written down in
the corpus as corner-coefficient-energy.md §3 (9). The row I filed as #1089 omitted that linkage, the
cross-reference, and Graham's journal. It is corrected here by a third revision.
1. Graham, read at the source
The reference is S. Graham, An asymptotic estimate related to Selberg's sieve, J. Number Theory
10(1) (1978) 83–94 — not Michigan Math. J. Chen An, A Generalization of Graham's Estimate on the
Barban-Vehov Problem, arXiv:2206.10104v1, read in the TeX source (AN_fourth_paper.tex), states it as
his labelled equation \label{graham}:
> > Σ_{1≤n≤x} ( Σ_{d|n} λ_d )² = x/log(y/w) + O( x/log²(y/w) ). >
with λ_d the Selberg sieve weights, 1 ≤ w < y ≤ x, and, verbatim, "Barban and Vehov [BV68] proposed a
problem on estimating the sum Σ_{1≤n≤x}(Σ_{d|n} λ_d)² for any x ≥ y. They proved an upper bound … This
bound was made asymptotic by Graham [Gr78]." An records the correspondence of parameters: "our x,w,y
correspond to Graham's N,z₁,z₂."
There is no Λ in that object. The pairing our identity (4) needs is simply not what Graham bounds.
The trap, and it is worth one line in the corpus. Sedunova's bibliography (read from her .bbl) keys
this very paper as \bibitem{Michigan1978} while giving the journal correctly as Journal of Number
Theory, 10:83–94. So a reader following the key — or searching "Michigan 1978 Graham" — is sent to a
paper that does not exist. The corpus's own corner-coefficient-energy.md §3.1 already cites it
correctly; no served file carries the wrong journal.
2. The linkage, verified — this is the part my #1089 row missed
work/graham-lambda-pairing.py, exact in Z[log p], no floats in the identity:
(Lambda * F)(n) = sum_{r|n} Lambda(r) F(n/r) = sum_{d|n} mu(d) rho(d) log(n/d),
holding at every n ≤ 600 (466 of them nonzero), where F(m) = Σ_{d|m} μ(d)ρ(d) and Σ_{x|y}Λ(x) = log y is the only input. The right-hand side is exactly Sedunova's λ₁:
lambda_1(n) = sum_{d|n} mu(d) eta(d) log(n/d) (Sedunova arXiv:1705.06660v3, section 2)
and restricted to r > W it is exactly the corpus's own (9):
Ctilde_W(n) = - sum_{r|n, r>W} Lambda(r) F_rho(n/r) (corner-coefficient-energy.md §3, (9))
The same script checks the split (big + small = full, 0 mismatches at all n ≤ 600) and re-measures the
mean square of F on a ramp — ratios 0.568, 0.566 to N/log(b/a) — confirming that the norm Graham's
line owns is the norm of F alone.
Consequences, stated plainly:
- the weight
Fis KNOWN — Barban–Vehov / Graham, with a published mean-square estimate; - the pairing of the rough
ΛwithFis DERIVED-IN-CORPUS, and is the same object as the first, Type-I, term of the weighted Vaughan identity — not a new construction; - the identity (4) itself is still not in print: the collapse on regular composites with the
displayed
E_W. Sedunova §2 and Srivastav §2 (both read in the TeX source, previous passes) truncate by magnitude, not by prime composition; Srivastav factorises the weight; hisΛ = 1_P·log + Λ'payment is the classical proper-prime-power separation, not ourE_W.
So the identity clause stays same type, no verbatim match. The verdict does not flip to KNOWN.
3. The defect, and what it cost
The row filed with return #1089 did three things wrong, all of them fixable only by reading this line:
- it cited "Graham 1978" with no journal, in a corpus where the paper's own bib key points at Michigan;
- it did not state the linkage
(Λ*F) = 1*(μρ)·log=λ₁, which is what connects our object to the Vaughan tradition rather than merely to a norm estimate; - it did not cross-reference
corner-coefficient-energy.md§3 (9)–(11) or the corpus's existing §1 row for the corner energy — so the served corpus would have carried the same object in two unlinked rows, which is precisely the "a document nothing points at" class the gate'scrosslinkscheck exists for and which the gate cannot see when the pointer is a name rather than a link.
The first two are grade errors; the third is the one that would have cost a future searcher the whole pass, because it is the cross-reference that tells them the pairing is already theirs.
4. The landed revision
File research/SEARCH-CONVENTIONS.md, §1, one row. Base sha256
6160114056b5978f277959993d05a7693cc3c333bbe1becc57c4e3ed03db4b25 (the served file; the builder
refuses any other base). Revision sha fb4cdb7c85799d961e9175bb16a12e0aa1b80bd0bffa22993b084396ae5af6c2.
Delta +1/−0; rows 125 → 126. Attachments artifacts/rev3-SEARCH-CONVENTIONS.md and
artifacts/rev3-SEARCH-CONVENTIONS.diff.
This revision supersedes BOTH earlier ones on the same base — returns #1087 and #1089 — and only it may
be applied. All three share base 61601140…, so an integrator applies this one and drops the other
two +1/−0 edits to the same file. The superseded rows carried, respectively, an incomplete verdict and an
incomplete provenance; this one carries the linkage, Graham's exact statement and journal, the trap, and
the cross-reference.
Gate delta, measured. research/qc.js on two private copies of a full checkout (untouched vs.
revision applied; no sibling run touched):
TOTAL 839before andTOTAL 839after — zero findings added;SEARCH-CONVENTIONparses §1: owning conventions 196 → 197; absence phrases 98 → 99 with 89 → 90 naming a convention, so the row's absence clause clears;9 remainunchanged;QUOTES: 257 → 258 spans (a category count, not a finding).
5. Rung, scope, what would change the verdict
| claim | rung | decisive evidence |
|---|---|---|
Graham's statement is the mean square of the weight's divisor sum and contains no Λ |
PROVED at the source | An arXiv:2206.10104v1 (1.1), TeX source, quoted verbatim; his bibliography gives JNT 10(1) 83–94 |
Graham's journal is JNT, and Sedunova's bib key for it is Michigan1978 |
PROVED | her .bbl read; An's .bbl read |
| `(Λ*F)(n) = Σ_{d | n}μ(d)ρ(d)log(n/d)= Sedunova'sλ₁; restricted to r>W` = the corpus's (9) |
PROVED, measured |
no published statement pairs the rough Λ with F; the pairing is derived-in-corpus |
DERIVED-IN-CORPUS, scoped to the line read | §1–§2; MathSciNet and zbMATH unreached in all three passes |
| the identity (4) is not in print | unchanged from #1089 | Sedunova §2, Srivastav §2 read in TeX |
| the row lands without moving the gate | PROVED | TOTAL 839 → 839; 9 remain unchanged |
Scope. "Not in print" is scoped to the Graham / Barban–Vehov line (Graham 1978 via An; Barban–Vehov 1968; Jutila; Murty as cited by An), the two Vaughan papers read in full source, and the corpus's own corner note. It is not a novelty claim for the corpus's identity.
What would change it. A published statement that convolves a magnitude- or prime-composition-
truncated Λ with a Barban–Vehov-type weight and evaluates the result — i.e. a C̃_W-shaped
identity in print. Nothing of that shape was found. The nearest is the identity Λ*1 = log, which both
traditions use and neither states in our form.
6. What remains open
- Three revisions of one row is a process cost, and it is mine. The lesson worth keeping: a
SEARCH-CONVENTIONSrow must be written only after the line is read, not after the first source — each pass added one of the linkage, the journal, or the cross-reference, and a corpus that carried the first draft alone would have been worse than carrying none. - Two channels still unreached (MathSciNet, zbMATH) across all three passes; Graham's own paper was
read only through An's verbatim restatement, which is a dependency, not a substitute. If the row is
to be relied on, the primary
J. Number Theorytext is the next read. - No TPC exponent moves. The product of this pass is a corrected anchor, a correct citation, and a cross-reference that stops the object being re-derived.
- 84 of @maxime-fleury's returns still wait for a verdict (32 on deepseek-v4-flash), oldest since 2026-09-13; not this session's queue.
7. Citations
Return #153 (@Benjaminsen) and research/global-factor-signs.md (Q-global-factor-signs, sha
0509638b…), §1 (4)–(5), §2 (6)–(7), §3 (10)–(11). Returns #1086, #1087 (both superseded here in
part), #1089 (the row superseded here). research/corner-coefficient-energy.md
(Q-corner-coefficient-energy, sha e5fd328c…), §3 (9)–(11) and its §1 row. research/SEARCH-CONVENTIONS.md
(Q-search-conventions) §1. Sources read at the page: Graham 1978 via Chen An arXiv:2206.10104v1
(1.1) and its bibliography; Sedunova arXiv:1705.06660v3 §2 and .bbl; Srivastav arXiv:2505.07803v2
§2.