=================================================================== --- a/paper/kk-lower-bound.md version 1 +++ b/paper/kk-lower-bound.md version 2 @@ -1,1239 +1,619 @@ -# A lower bound for the two-class Jacobsthal function +# A two-class Erdos and Rankin lower bound for the Jacobsthal gap -*Draft, 2026-08-28. Internal to the primeoire repository; the publication -moratorium is in force and this document is not submission copy. Prose follows -`paper/writing-style-math.md`. Grade and triggers are carried by -`paper/proposals/prop-kk-lower-bound.md`; the suite architecture is -`paper/PAPERS.md`.* +**Author:** Chris Benjaminsen. -**Author.** Chris Benjaminsen. +**Status:** Draft for project review, 18 September 2026. This manuscript +implements the proposal `kk-lower-bound`. Its main composition is +**DERIVED / INFERRED, not independently refereed**, in the terminology of the +project's proposal registry. It is not a journal or arXiv submission and does +not lift the publication moratorium recorded in `paper/PAPERS.md`. -**Methods and AI disclosure** (the statement adopted for the suite, -`paper/PAPERS.md`, adapted to this note): the framework, vocabulary, and -driving questions are the author's, developed over six years of independent -work. Formal derivations, literature audits, computations, and manuscript -drafting were carried out using an AI assistant operating -under the author's direction; all results were verified by explicit -computation, with code and outputs published in the accompanying repository, -and all refuted intermediate claims retained in the record. - ---- - ## Abstract -Let $P(y) = \prod_{p \le y} p$ and let $G_2(N)$ be the largest gap between -consecutive integers $n$ with $\gcd(n(n+2), N) = 1$, the two-class analogue of -Jacobsthal's function $g$. The main statement of this note is +This argument proves no assertion about the infinitude of twin primes. It +constructs long intervals without integers that survive the two residue +exclusions modulo every prime up to a given threshold. Write +$P(y)=\prod_{p\leq y}p$, and let $G_2(P(y))$ be the largest gap between +successive integers $n$ satisfying $\gcd(n(n+2),P(y))=1$. We give the +project's derivation of +$$ + G_2(P(y))\gg + \frac{y(\log y)^3(\log\log\log y)^2}{(\log\log y)^4}. + \tag{1} +$$ +The argument adapts the three-band construction of Kalmynin and Konyagin, +rather than applying their polynomial theorem to a different object. +The imported analytic inputs are their upper-sieve Lemma 1, Mertens' +theorem, the prime number theorem, and the smooth-number estimate of +Hildebrand and Tenenbaum, Corollary 1.3. We state these dependencies where +they enter and prove the intervening covering and counting steps. +The source's 1974 sieve-theorem reference has not been checked at the printed +page in this work. No implied constant or numerical starting threshold is +certified. Equation (1) remains a project-derived, unrefereed composition +from the stated inputs. -$$G_2(P(y)) \;\gg\; \frac{y (\ln y)^3 (\ln\ln\ln y)^2}{(\ln\ln y)^4} -\qquad (y \ge y_0),$$ +## 1. Object, contribution and prior work -with an absolute implied constant, obtained by substituting the kill set -$\Omega_p = \{a_p, a_p - 2\}$ into the multi-class Erdős and Rankin -construction published by Kalmynin and Konyagin, and carrying their §2 -trichotomy through line by line. The construction is theirs; the substitution -and the proof that it goes through are the present author's, and the argument -follows the source's construction while proving each step here, consuming the -source's Corollary 1 at its statement. It has been checked twice adversarially -inside this project and it has **not been refereed**. +Let +$$ + {\cal T}_y=\{n\in\mathbb Z:\gcd(n(n+2),P(y))=1\}. +$$ +This set is periodic with period $P(y)$ and is nonempty: $-1$ belongs +to it. We define $G_2(P(y))$ as the largest distance between consecutive +members of this periodic set, including the gap across a period boundary. +A gap of length $G$ therefore contains $G-1$ excluded integers. +All logarithms in this note are natural. -Two weaker statements are recorded beside it, and both are stronger in -provenance than in exponent. The free bound $G_2 \ge g$ imports the -one-class Erdős and Rankin literature at no cost and gives -$G_2(P(y)) \gg y \ln y \ln\ln\ln y / \ln\ln y$. A chain assembled from -published statements alone, Kalmynin and Konyagin's Corollary 1 together with -Mertens, the prime number theorem, and an elementary identity from the Chinese -remainder theorem, gives $G_2(P(y)) \gg y \ln y$ with an effective constant. -The three sit two logarithms apart at the top and one $\ln\ln y / \ln\ln\ln y$ -apart at the bottom. +The object predates the project's terminology. Under the translation +$t=n+1$, exclusion means $t\equiv1$ or $-1\pmod p$ for some prime +$p\leq y$. Consequently, at the $k$-th prime $p_k$, +$$ + G_2(P(p_k))-1=\mathrm{A144311}(k). + \tag{2} +$$ +The current OEIS entry attributes the sequence to Andrew Carter, with +extensions by Max Alekseyev and Jinyuan Wang [O]. Identity (2) concerns +the definition, not a new computation of its entries. -Nothing here touches the twin prime conjecture, and nothing here moves the -standing upper bound $G_2 \ll_\varepsilon y^{\,\beta_2 + \varepsilon}$ at -$\beta_2 = 4.266$. The gap between the two sides remains a factor -$y^{3.266 + o(1)}$ wide. +The construction used below belongs to Kalmynin and Konyagin [KK]. +Their polynomial quantity asks for a translate $b+f(i)$ that fails +coprimality for every $1\leq i\leq m$. At $f(i)=i(i+2)$, its local +excluded set solves +$$ + i(i+2)+b_p=0\pmod p. +$$ +When the two roots exist at an odd prime, they have centre $-1$ and +separation depending on $b_p$. Our excluded pair +$\{a_p,a_p-2\}$ has a moving centre and a fixed separation. Thus [KK, +Theorem 1] is not itself a theorem about $G_2$. Sections 4--7 below +adapt its proof architecture to this specific pair. ---- +The project already records this adaptation in +`research/two-class-lower-bounds.md`, section 4c, and in the live manuscript +`paper/kk-lower-bound.md`, sections 4--7 [R2, R4]. This draft supplies a +compact exposition, retains the later source corrections, and gives an +incidence-weight derivation of the residue-class sieve estimate directly +from [KK, Lemma 1]. It makes no claim to a new multi-class construction or +to priority for the two-class specialization. -## 1. Introduction +The source registry distinguishes this fixed-offset quantity from paired +Jacobsthal functions optimized over all even offsets and from systems with +two independently chosen classes per prime [R5]. Neither comparison +identifies these different extremal problems. -### 1.1 One class per prime +## 2. The exact covering identity -Jacobsthal's function $g(N)$ is the largest gap between consecutive integers -coprime to $N$. At a primorial $N = P(y)$ it is the object of Erdős problem -#687, which carries a US\$1000 prize, and its history is the history of two -opposed techniques. +**Proposition 1 (elementary; [R2, section 1]).** The largest integer $m$ +for which one can cover $\{1,\ldots,m\}$ by pairs +$$ + \{a_p,a_p-2\}\pmod p,\qquad p\leq y, + \tag{3} +$$ +equals $G_2(P(y))-1$. -On the upper side, Iwaniec proved $g(P(y)) \ll y^2$, equivalently -$g(N) \ll (\ln N)^2$, in 1978 [Iw78]. That bound is the linear sieve run at its -sifting limit $\beta_1 = 2$, and the numerical agreement between the sifting -limit and the exponent 2 is what makes it land where it does. Nothing better is -published, and the only proof of the transfer step, Iwaniec's Lemma 1, has an -unanswered question mark against it on MathOverflow that has stood since 2016 -(MO 245539). This note does not touch the upper side. +**Proof.** Given the choices $a_p$, the Chinese remainder theorem +provides an integer $s$ satisfying $s\equiv-a_p\pmod p$ for every +prime $p\leq y$. For an index $i$, membership in the pair (3) says +that $p\mid s+i$ or $p\mid s+i+2$. A covered interval of $m$ +indices therefore gives $m$ consecutive integers outside ${\cal T}_y$. +Since ${\cal T}_y$ is nonempty and periodic, this interval lies between +two successive members at distance at least $m+1$. -On the lower side, the technique is a covering construction. Erdős and Rankin -choose one residue class $a_p$ per prime and ask how long an interval the -chosen classes can cover; the covered interval, transported by the Chinese -remainder theorem, is a run with no coprime integer in it. Rankin's 1938 -argument, Pintz's 1997 refinement, and the Ford, Green, Konyagin, Maynard and -Tao theorem of 2018 give +Conversely, let $s$ be the left endpoint of a gap of length $G$ in +${\cal T}_y$. Each $s+i$, $1\leq i 1$ for every $i \le m$, and they prove a lower -bound whose exponent is governed by two parameters of $f$: the number -$\ell_f$ of linear factors, and $M(f)$, the average size of the maximal -preimage of a point under $f : \mathbb{F}_p \to \mathbb{F}_p$. Each prime now -deletes a whole fibre rather than a single class, and the Erdős and Rankin gain -is raised to the power $M(f)$. +We isolate the upper-sieve estimate because both lower-bound constructions +in this note depend on it. -### 1.2 Two classes per prime +**Input S ([KK, Lemma 1]).** Let $\kappa$ be fixed. For primes $p\leq z$, +suppose a multiplicative function $g$ satisfies +$0\leq g(p)\leq\kappa$ and $g(p) 1 \text{ for } a < n < b \,\},$$ +**Input H ([HT, Corollary 1.3, printed page 417]).** With +$\Psi(X,Z)$ counting integers at most $X$ all of whose prime factors +are at most $Z$, put $u=\log X/\log Z$. For a fixed +$\varepsilon>0$, +$$ + \Psi(X,Z)=X u^{-(1+o(1))u} + \tag{9} +$$ +as $Z,u\to\infty$, uniformly when $u\leq Z^{1-\varepsilon}$. +The formula and its range were read from the primary page image. +We will use only its upper-bound consequence. -the largest gap between consecutive twin-admissible slots modulo $N$. It is not -a new object. Shifted by one, $G_2(P(p_n)) - 1$ is **OEIS A144311**, entered by -Andrew Carter in September 2008 and carrying 22 terms to $y = 79$, extended by -Alekseyev in 2009 and by Wang in 2024, whose entry words the same quantity as -*"the length of the longest sequence of consecutive integers, each equal to 1 -or $-1$ modulo at least one of the first $n$ primes"*. A144311 carries no -formula line and no reference line. The related paired Jacobsthal function -$h_2$ of Ziller and Morack, OEIS A288815, takes the worst case over all even -offsets and satisfies $h_2 \ge G_2$; their papers give a conjectural ceiling -and no lower bound. +**Input P.** The prime number theorem gives +$$ + \pi(y)-\pi(y/2)\sim\frac{y}{2\log y}. + \tag{10} +$$ +No explicit prime-count threshold is required. This is the final counting +input in [KK, section 2] and [R4, section 7]. -Two facts orient the problem. +## 4. A residue-class form of Input S -**The upper side has one published-ingredient bound and it is far away.** The -dimension-2 lower-bound sieve of Diamond, Halberstam and Richert has sifting -limit $\beta_2 = 4.266$ as printed on p. 79 of their tract, and running -Iwaniec's argument with it gives $G_2(P(y)) \ll_\varepsilon -y^{\,\beta_2 + \varepsilon}$. That statement is derived inside this project -(`paper/beta2-note.md`) with its sieve input checked line by line against the -primary source, and it is the only two-class upper bound at any exponent that -the searches recorded in `research/SEARCH-CONVENTIONS.md` §3 could find. +**Lemma 2 (elementary reduction from Input S).** For sets +$\Omega_p\subset\mathbb Z/p\mathbb Z$ with +$g(p)=|\Omega_p|\leq\kappa$ and $g(p)X$. -### 1.3 What this note does +Since $D(n)\mid P(z)$, it is coprime to $P(z)$ exactly when +$D(n)=1$. Hence the left side of (7) is exactly $S(X,\Omega)$. +Applying Input S proves (11). $\square$ -Kalmynin and Konyagin's construction is not a theorem about $G_2$ and may not -be evaluated as one. At $f(x) = x(x+2)$ their deleted fibre is -$\{-1 \pm \sqrt{1 - x_p}\}$, a pair with fixed centre and varying separation; -the pair that defines $G_2$ has varying centre and separation fixed at 2. The -two families coincide only at $x_p = 0$, where both are $\{0, -2\}$. So their -Theorem 1 cannot be read off. What transfers is the construction, instantiated -on a different system, and the result is a different proof of a different -theorem that lands on the same right-hand side. +This is the conclusion of [KK, Corollary 1], expressed through counting +weights instead of its printed polynomial encoding. The live source +record reports a representative and sign problem with that encoding +[R4, section 11.2; R6]. The proof above does not use it or an unstated +choice of integer representatives. The remaining imported obligation is +Input S itself. No Selberg support parameter or additional remainder sum +is introduced into (11). -Section 4 sets up that instantiation, §5 re-derives their §2 trichotomy for -$\Omega_p = \{a_p, a_p - 2\}$, §6 runs the sieve and the Mertens ledger, and §7 -finishes. Section 8 gives the explicit threshold and says exactly what it is a -threshold for. Section 9 proves the weaker $y \ln y$ bound from published -statements only, which is the version a reader unwilling to grant a reading of -someone else's proof can still take. Section 10 states what none of this does. +## 5. The three bands and the finite covering step -### 1.4 Attribution +Fix a constant $A>4$. A second constant $B>0$ will be chosen after +the sieve constant is accounted for. Write +$$ + L=\log y,\qquad L_2=\log\log y,\qquad L_3=\log\log\log y, +$$ +and set +$$ + z_0=L^A,\qquad + z_1=\exp\!\left(\frac{L L_3}{A L_2}\right),\qquad + m=\left\lfloor\frac{y}{B}\frac{L^3L_3^2}{L_2^4}\right\rfloor . + \tag{13} +$$ +These are the parameters of [KK, section 2], instantiated as in +[R2, section 4c; R4, section 4]. -The multi-class Erdős and Rankin construction is Kalmynin and Konyagin's, and -the record says so before it says anything else -(`research/SEARCH-CONVENTIONS.md` §3, `research/PRIOR-ART.md`). What is claimed -here is the two-class instantiation and the carry-through, and the -carry-through consumes their §2 while discarding the whole of their §3. +For fixed $A,B$, all sufficiently large $y$ satisfy +$$ + 34$. +We assert eventual validity, not a certified numerical value of $y$. -The framing this note sits inside, primorial coordinates for the sieve wheel -and its recursive structure, has a predecessor that owns most of it: Fred B. -Holt, with Helgi Rudd on the earlier papers, has run a programme since 2007 on -the cycle of gaps $G(p\#)$ among the generators of $\mathbb{Z}/p\#$, with the -fold recursion, the closure theorem, the transfer operator and the interval of -survival all in print (arXiv:1408.6002, arXiv:2603.25915, primegaps.info). The -present project reached the same structures independently and cites them where -it uses them, per `research/PRIOR-ART.md`. Holt's corpus contains no upper -bound on any maximum gap and no two-class covering construction, so the object -of this note is not in it; the frame around the object is. +Choose the covering classes as follows: -The standing posture of this project's prior-art record applies here without -softening. Its central object sat in OEIS from 2008 while five internal audit -waves ran clean negatives in the wrong vocabulary. A negative here is worth the -search behind it and no more, and §11.3 states which conventions were searched -and which have no row yet. - ---- - -## 2. The object as a covering problem - -Write $P(y) = \prod_{p \le y} p$ throughout, and call $n$ a *twin slot* modulo -$N$ when $\gcd(n(n+2), N) = 1$. The primorial $P(y) - 1$ is a twin slot, so -twin slots exist at every level and $G_2$ is well defined. - -> **Proposition 1 (elementary, Chinese remainder theorem).** -> $G_2(P(y)) - 1$ equals the largest $m$ for which there is a choice of one -> residue $a_p \in \mathbb{Z}/p\mathbb{Z}$ for each prime $p \le y$ such that -> every $i \in [1, m]$ satisfies $i \equiv a_p$ or $i \equiv a_p - 2 \pmod p$ -> for at least one $p \le y$. - -*Proof.* A gap of length $G$ starting at a twin slot $s$ says that each of -$s+1, \dots, s+G-1$ fails to be a twin slot, that is, for each such $t$ either -$t$ or $t+2$ shares a factor with $P(y)$. Translate the origin to $s$ and write -$i = t - s$. Then $p \mid s + i$ exactly when $i \equiv -s$, and -$p \mid s + i + 2$ exactly when $i \equiv -s - 2$, so the deleted pair is -$\{a_p, a_p - 2\}$ with $a_p = -s \bmod p$. As $s$ runs over -$\mathbb{Z}/P(y)$, the Chinese remainder theorem makes $(a_p)_{p \le y}$ run -over all of $\prod_{p \le y} \mathbb{Z}/p\mathbb{Z}$. Conversely a covered run -is bounded on both sides by uncovered points, and those are twin slots. $\square$ - -So $G_2$ has an adversary after all, and the adversary's only constraint is -that the two deleted classes sit at distance exactly 2. Everything below is a -construction for that adversary. The identity is not new to this project's -record either: `research/attack2-rankin2d.js` already carried it under the name -PAIRED, with the note that choosing all shifts independently is the same as -sliding one window over the tile. - -Three neighbouring quantities, all pointwise comparable: - -| object | classes deleted per prime | relation | OEIS | -|---|---|---|---| -| $g(P(y))$, Jacobsthal | one, $\{a_p\}$ | base | A048670 | -| $G_2(P(y))$ | $\{a_p, a_p - 2\}$ | $\ge g(P(y))$ | A144311 (shifted by 1) | -| $h_2$, paired Jacobsthal | $\{a_p, a_p - d\}$, worst even $d$ | $\ge G_2(P(y))$ | A288815 | -| free two-class | any $\{a_p, b_p\}$ | $\ge h_2$ | A072753 | - -The inequality $G_2 \ge g$ is verified at all eleven levels where both ladders -are known, with the ratio $G_2/g$ running 2.00, 3.00, 3.00, 3.00, 4.15, 4.41, -5.10, 5.61, 6.00, 8.00, 7.38 for $y = 5$ to $41$ -(`research/two-class-lower-bounds.md` §1). - ---- - -## 3. Statements - -Three bounds, in increasing strength and decreasing provenance. - -> **Proposition 2 (proven, and free).** -> $$G_2(P(y)) \;\ge\; g(P(y)) \;\gg\; \frac{y \ln y \ln\ln\ln y}{\ln\ln y}.$$ -> The first inequality is pointwise and elementary (§2); the second is Ford, -> Green, Konyagin, Maynard and Tao, *JAMS* **31** (2018), via Rankin 1938 and -> Pintz 1997. - -> **Theorem A (proven from published statements; the composition is the -> author's, adversary-confirmed, not refereed).** -> There is an effective $c > 0$ with -> $$G_2(P(y)) \;\ge\; (c + o(1))\, y \ln y.$$ -> Every ingredient is a published theorem consumed at its statement: -> Kalmynin and Konyagin's Corollary 1 at sieve dimension $\kappa = 2$, Mertens' -> third theorem, the prime number theorem, and Proposition 1. Proof in §9. - -> **Theorem B (derived here from a published construction; not refereed).** -> There is an absolute $y_0$ such that for every $y \ge y_0$, -> $$G_2(P(y)) \;\gg\; \frac{y (\ln y)^3 (\ln\ln\ln y)^2}{(\ln\ln y)^4},$$ -> with an absolute implied constant. The proof substitutes -> $\Omega_p = \{a_p, a_p - 2\}$ into Kalmynin and Konyagin's construction and -> carries their §2 trichotomy through at $\ell_f = 2$, $h_f = 0$, $M(f) = 2$. -> Proof in §§4 to 7; the threshold in §8. - -**Calibration, stated in the project's own legend -(`paper/proposals/PROPOSALS.md`).** Proposition 2 is PROVEN, and both of its -halves are in print. Theorem A is PROVEN at every ingredient and INFERRED at -the composition: the composition is written out in full here and was -re-derived end to end by a dedicated adversarial pass -(`research/history/staging/redteam-0820-math.md` §3.3), and no referee has seen -it. Theorem B is INFERRED at the composition, like Theorem A, and its sieve -step consumes the same printed statement (Corollary 1). It sits below Theorem -A on three counts: one ingredient, the smooth-number estimate of §6.5, is -consumed without an artifact (§12, [MEMORY]); its threshold is a floor with -every implied constant set to 1 (§8); and its construction cannot be -exhibited at any computable scale (§10). It must never be quoted without the -derivation named. - -**Placement.** Writing $\ell = \ln y$, $\ell\ell = \ln\ln y$, -$\ell\ell\ell = \ln\ln\ln y$: - -| bound | right-hand side | ratio to the one below | +| Primes | Choice of $a_p$ | Pair excluded modulo $p$ | |---|---|---| -| Theorem B | $y \ell^3 (\ell\ell\ell)^2 / (\ell\ell)^4$ | $\approx \ell^2 (\ell\ell\ell)^2/(\ell\ell)^4$ | -| Theorem A | $y \ell$ | $\ell\ell / \ell\ell\ell \to \infty$ | -| Proposition 2 | $y \ell \, \ell\ell\ell / \ell\ell$ | base | +| $p\leq z_0$ or $z_1 4$ and $B$ large in terms of $A$, and set +In the notation of [KK], no nonlinear-factor case remains: the two +forms $i$ and $i+2$ are linear. There is no Chebotarev or +Galois-group input to Proposition 3 or to the argument below. -$$z_0 = (\ln y)^A, \qquad -z_1 = \exp\!\left(\frac{\ln\ln\ln y \cdot \ln y}{A \ln\ln y}\right), \qquad -m = \frac{y}{B} \cdot \frac{(\ln y)^3 (\ln\ln\ln y)^2}{(\ln\ln y)^4}.$$ +## 6. Counting the uncovered indices -These are Kalmynin and Konyagin's $z_0$, $z_1$ and $m$, the last evaluated at -$\ell_f = 2$, $h_f = 0$, $M(f) = 2$. The displays were read from the arXiv -PDF of record at pp. 3 and 6 and checked against the published Izvestiya PDF -(§12 and Appendix A). - -Choose one residue $a_p$ per prime $p \le y$ in three bands. - -| band | primes | $a_p$ | deleted pair $\{a_p, a_p - 2\}$ | -|---|---|---|---| -| 1 | $p \le z_0$ and $z_1 < p < y/2$ | $0$ | $\{0, -2\}$ | -| 2 | $z_0 < p \le z_1$ | $1$ | $\{1, -1\}$ | -| 3 | $y/2 \le p \le y$ | greedy, one leftover $i$ per prime | one point | - -By Proposition 1 it suffices to show that every $i \in [1, m]$ is deleted by -some prime, since then $G_2(P(y)) \ge m + 1$, which is the right-hand side of -Theorem B up to the constant $1/B$. - -The sieve sets fed to the fundamental lemma at sifting level $z = \sqrt{y}$ are - -$$\Omega^{\mathrm{I}}_p = \{0, -2\} \ \ (p \le \sqrt{y}), \qquad -\Omega^{\mathrm{II}}_p = \emptyset, \qquad -\Omega^{\mathrm{III}}_p = \{1, -1\} \ \ (z_0 < p \le z_1),$$ - -and $\Omega_p$ is their union. The second is empty because $f(x) = x(x+2)$ has -no irreducible non-linear factor over $\mathbb{Z}$, so $h_f = 0$; §5 makes that -the content of Case 2. - -**Why the two formulations never have to be reconciled.** Proposition 1's -covering pair $\{a_p, a_p - 2\}$ has a free centre; the sieve pair -$\Omega^{\mathrm{I}}_p = \{0, -2\}$ is fixed. Band 1 sets $a_p = 0$, and at -$a_p = 0$ the two sets are equal. The covering freedom is therefore spent -entirely in band 2, where it is spent on the single value $a_p = 1$, and each -formulation is used only in the band where it applies. This project has made -the covering-versus-sifting error before, which is why `research/qc/units.js` -§5 exists, and it is why the point is made here rather than passed over. - -**Three structural facts the substitution needs, each a resultant rather than a -sweep.** - -| what the method needs | what the substituted system gives | reason | -|---|---|---| -| $\lvert \Omega^{\mathrm{I}}_p\rvert = \ell_f$ above some $p_0$ | $\lvert \Omega^{\mathrm{I}}_p\rvert = 2$ for every $p > 2$ | $\mathrm{Res}(x, x+2) = 2$ | -| the three $\Omega$ pairwise disjoint above $p_0$ | disjoint for every $p \ge 5$; $p = 3$ is the last overlap | $\mathrm{Res}(x(x+2), (x-1)(x+1)) = -3$ | -| a sieve dimension $\kappa$ (theirs is $3 \deg f = 6$) | $\kappa = 4$ | $\lvert \Omega^{\mathrm{I}}\rvert + \lvert \Omega^{\mathrm{III}}\rvert = 4$, $\Omega^{\mathrm{II}} = \emptyset$ | - -The band-2 choice $a_p = 1$ is optimal in the sense that matters here: -$\mathrm{Res}(x(x+2), (x-a)(x-a+2)) = a^2(a^2 - 4)$, the values -$a \in \{0, 2, -2\}$ are degenerate, and among the rest $a = \pm 1$ minimise -$|\mathrm{Res}|$ at 3, which is the floor because $3 \mid a^2(a^2-4)$ whenever -$3 \nmid a$. - -The count $|\Omega^{\mathrm{III}}_p| = 2$ was also checked directly at every -one of the 78,498 primes below $10^6$, with no exception -(`research/attack-kk-substitution.js`). That check is redundant given the -resultant, and it is recorded because it is the form the first adversarial pass -demanded. - ---- - -## 5. The trichotomy - -Throughout this section $i \in [1, m]$, and *unkilled* means unkilled by bands -1 and 2. - -> **Proposition 3.** If $i \le m$ is unkilled after bands 1 and 2, then at -> least one of the following holds: -> **(a)** $i \le \sqrt{y} + 2$; -> **(b)** $i$ or $i+2$ is $z_1$-smooth; -> **(c)** $i \bmod p \notin \Omega_p$ for every $p \le \sqrt{y}$. - -*Proof.* Suppose (c) fails, so $i \bmod p \in \Omega_p$ for some -$p \le \sqrt{y}$. Since $\Omega_p$ is the union of the three sets of §4, one of -three cases holds. - -**Case 1: $i \bmod p \in \Omega^{\mathrm{I}}_p$.** Then $i \equiv 0$ or -$i \equiv -2 \pmod p$, that is $p \mid k$ with $k = i$ or $k = i+2$, so -$1 \le k \le m+2$. If $k = p$ then $i \le p + 2 \le \sqrt{y} + 2$ and (a) -holds. Otherwise $k$ is composite. Since $i$ is unkilled by band 1, neither $i$ -nor $i+2$ has a prime factor in $[2, z_0] \cup (z_1, y/2)$, so every prime -factor of $k$ lies in $(z_0, z_1] \cup [y/2, \infty)$. Let $P$ be the largest. -If $P \le z_1$ then $k$ is $z_1$-smooth and (b) holds. If $P \ge y/2$ then -$k/P \le (m+2)/(y/2) = 2(m+2)/y$, while every prime factor of $k/P$ exceeds -$z_0$, so that $k/P > z_0$ unless $k/P = 1$. Under the inequality - -$$2(m+2)/y \le z_0 \tag{5.1}$$ - -the cofactor $k/P$ therefore has no prime factor at all. Hence $k/P = 1$ and -$k = P$ is prime, contradicting that $k$ is composite. (Until 2026-09-07 the -first display read $(m+2)/(y/2) < 2m/y$, which is false by $4/y$, and (5.1) -read $2m/y < z_0$; the repair changes H4 by $\ln(1 + 2/m)$, below $10^{-100}$ -at every tabulated $L$, and no brute-force row: `research/history/reviews-0907/09`.) - -**Case 2: $i \bmod p \in \Omega^{\mathrm{II}}_p$.** The source defines -$\Omega^{\mathrm{II}}_p$ as $\{t : q(t) \equiv 0 \pmod p$ for some non-linear -irreducible factor $q$ of $f\}$, read from the PDF of record at p. 6. Since -$f(x) = x(x+2)$ is a product of two linear factors over $\mathbb{Z}$, -$h_f = 0$, the set $\Omega^{\mathrm{II}}_p$ is empty at every prime, and this -case quantifies over an empty set. It is unsatisfiable and contributes nothing. - -**Case 3: $i \bmod p \in \Omega^{\mathrm{III}}_p$.** Then $z_0 < p \le z_1$ and -$i \equiv 1$ or $i \equiv -1 \pmod p$, which is exactly the band-2 deleted -pair, so $i$ was killed at band 2, contrary to assumption. $\square$ - -**What Case 1 consumes, and it is all of what it consumes**, is the single -inequality (5.1). Since $m \le y (\ln y)^3$, that step alone asks only -$A > 3$. The proof as a whole needs $A > 4$, and it needs it at the -smooth-number step of §6, not here. An earlier version of this derivation wrote -$A > 3$ for the whole proof and the first adversarial pass corrected it -(`research/history/staging/verify-kk-substitution.md`). - -**What Case 2 takes with it.** Kalmynin and Konyagin's Lemma 2, their use of -the effective Chebotarev density theorem, their Theorem 2, their Lemma 3 and -the Galois groups $G_f$ and $G_f^+$, which is the whole of their §3, are -consumed by nothing here. Lemma 2 and Chebotarev drop because -$\Omega^{\mathrm{II}}$ is empty. Theorem 2, Lemma 3 and the Galois groups drop -for the Case 3 reason instead: they exist in order to compute $M(f)$, and the -substituted $\Omega^{\mathrm{III}}_p$ has exactly two elements at every odd -prime, with no logarithmic average and no exceptional primes. About half of the -source paper is not needed. - -**The finite content of Proposition 3 was brute-forced.** At $y = 200000$, -$z_1 = 300$, $m = 4 \cdot 10^6$, so the Case 1 threshold (5.1) reads -$z_0 \ge 2(m+2)/y = 40.00002$ (the script tests $2m/y = 40$; no tabulated -$z_0$ lies between): - -| $z_0$ | does (5.1) hold | unkilled $i \le m$ | counterexamples to Proposition 3 | -|---|---|---|---| -| 100 | yes | 16985 | 0 | -| 60 | yes | 14381 | 0 | -| 45 | yes | 13164 | 0 | -| 30 | no | 11228 | 235 | -| 20 | no | 10372 | 799 | -| 10 | no | 9086 | 2921 | - -The correct reading is that (5.1) is **sufficient** at finite scale, -counterexample-free wherever it holds, and that violating it far enough breaks -the proposition. It is not necessary: violations near the boundary produce no -counterexamples, measured across three parameter families. The run also rules -out a covering-versus-sifting leak of the kind §4 warns about, because such a -leak would surface as a counterexample where (5.1) holds. The producer is -`research/verify-kk-substitution.js`; the report is -`research/history/staging/verify-kk-substitution.md`. - ---- - -## 6. The sieve step and the Mertens ledger - -Let $R$ be the number of $i \le m$ unkilled after bands 1 and 2. Proposition 3 +Let $R$ be their number after the first two covering bands. Proposition 3 gives +$$ + R\leq S(m,\Omega)+O(\sqrt y)+2\Psi(m+2,z_1). + \tag{17} +$$ +The smooth-number terms may overlap; an upper bound is all that is needed. -$$R \;\le\; S(m, \Omega) \;+\; O(\sqrt{y}) \;+\; 2\,\Psi(m+2, z_1), \tag{6.1}$$ +### 6.1 The sieve conditions -where $S(X, \Omega) = \#\{n \le X : n \bmod p \notin \Omega_p \ \forall p \le -z\}$ and $\Psi$ is the smooth-number counting function. The factor 2 on the -smooth count is there because branch (b) of Proposition 3 is a statement about -$i$ or $i+2$. +At $p=2$ the basic pair has one class, and at $p=3$ it has two. +Both primes are below $z_0$. At primes $p\geq5$, the four residues +in the middle-band set are distinct: a collision between +$\{0,-2\}$ and $\{1,-1\}$ would force $p\mid1$ or $p\mid3$. +Thus $g(2)=1$, $g(3)=2$, and $g(p)$ is 2 or 4 thereafter. +In every case +$$ + 0\leq g(p)\leq4,\qquad g(p) *Let $\kappa$, $z$, $g(d)$, $V(z)$ and $X$ be as above. Suppose that for any -> $p \le z$ the set $\Omega_p \subset \mathbb{Z}/p\mathbb{Z}$ contains $g(p)$ -> elements. Let $S(X, \Omega)$ be the number of $n \le X$ such that -> $n \bmod p \notin \Omega_p$ for all $p \le z$. Then $S(X, \Omega) \ll X V(z)$.* +Multiplying (21) by (13) gives +$$ + S(m,\Omega)\ll \frac{A^4}{B}\frac{y}{L}. + \tag{22} +$$ +After fixing $A$, choose $B$ large enough that this term is at most +$y/(4L)$ for all sufficiently large $y$. We have not assigned a +numerical value to the sieve constant. -Here $V(z) = \prod_{p \le z} (1 - g(p)/p)$, and "as above" imports Lemma 1's -hypotheses: $g$ multiplicative, $g(p) \le \kappa$, $g(p) < p$ for all primes, -$|r_d| \le g(d)$ for every $d \mid P(z)$, $z \ll X$, and an implied constant -depending on $\kappa$ alone. +### 6.3 The smooth-number term -**The corollary sees $\Omega_p$ only through $|\Omega_p|$, and that single fact -is what makes the substitution legal**, because it is indifferent to whether -the classes are a fibre of a polynomial or a free translate. Their own proof -invokes it with the words *"Let $g(p) = |\Omega_p|$"*. This reading is -load-bearing and §11.1 flags it as such. +Set $X=m+2$, $Z=z_1$. Then +$$ + \log X=L+O(L_2),\qquad + u=\frac{\log X}{\log Z}\sim\frac{A L_2}{L_3}. + \tag{23} +$$ +Both $u$ and $Z$ tend to infinity. For example, with +$\varepsilon=1/2$, the range condition $u\leq Z^{1-\varepsilon}$ +holds eventually, because +$\log u=O(L_3)$ whereas +$\log Z=L L_3/(A L_2)$. -The hypotheses are discharged as follows. $g(p) = |\Omega_p| \le 4$ everywhere, -with 4 attained only on band 2, so $\kappa = 4$. The condition $g(p) < p$ holds -for $p \ge 5$, and the two primes where a band-2 count would fail it, $p = 2$ -and $p = 3$, sit in band 1 where $g(2) = 1$ and $g(3) = 2$, because band 2 -begins above $z_0 = (\ln y)^A > 3$. On squarefree $d$ put $g(d) = \prod_{p -\mid d} g(p)$; then, by the Chinese remainder theorem, the $n \le m$ with $n -\bmod p \in \Omega_p$ for every $p \mid d$ fill exactly $g(d)$ classes mod $d$, -each containing $m/d + O(1)$ integers, so the remainder satisfies $|r_d| \le -g(d)$ for every $d \mid P(z)$ with no restriction on the size of $d$. Finally -$z = \sqrt{y} \ll m$, since $m/\sqrt{y} \to \infty$ for fixed $B$. A smaller -$\kappa$ than the source's own $\kappa = 6$ is a tightening rather than a -weakening, since $\kappa$ enters only the implied constant. +Moreover $u\log u=(A+o(1))L_2$. Input H therefore implies +$$ + \Psi(m+2,z_1) + =(m+2)L^{-A+o(1)} + =o(y/L)\qquad(A>4). + \tag{24} +$$ +For the last step, the ratio to $y/L$ is bounded by a constant times +$L^{4-A+o(1)}L_3^2/L_2^4$, which tends to zero. +This discharges the smooth-number range rather than assuming that a +fixed-$u$ estimate applies while $u$ grows. -### 6.2 Remark: no support parameter is consumed +Also $\sqrt y=o(y/L)$. Combining (17), (22) and (24), for sufficiently +large $y$, +$$ + R\leq\frac{y}{3L}. + \tag{25} +$$ -*(Demoted from an argument to a remark on 2026-09-07 after a source re-read; -see §11.2 and `research/history/reviews-0907/05`, `06`.)* Lemma 1 of the -source is the Brun-form upper bound: under $g(p) \le \kappa$, $g(p) < p$, the -pointwise remainder hypothesis $|r_d| \le g(d)$ for $d \mid P(z)$, and $z \ll -m$, it gives $S \ll_\kappa m V(z)$ with no remainder sum and no support -parameter. Richert's own cross-reference identifies it with Halberstam and -Richert Theorem 2.2 (Tata lectures, Theorem 11.3: $S \ll X \prod_{p 3\kappa/2$, -sieving at $z = \xi$, at a cost $1 + o(1)$ in $V$; $C = 7$ serves $\kappa = 4$ -and $C = 10$ serves $\kappa = 6$, so nothing about it is one sieve dimension -wide. It is not consumed anywhere below: every computation in §§6.3 to 8 runs -at $z = \sqrt{y}$, as the earlier text of this subsection did not. +By Input P, the reserved interval $[y/2,y]$ contains more than +$y/(3L)$ primes for all sufficiently large $y$. Assign distinct +reserved primes $p_i$ to the $R$ remaining indices. Set +$a_{p_i}=i\pmod{p_i}$. This covers index $i$; covering other indices +as well causes no difficulty. Choose any value for unused reserved +primes. -### 6.3 The ledger has exactly two terms +All indices $1,\ldots,m$ are now covered by admissible pairs. +Proposition 1 gives +$$ + G_2(P(y))\geq m+1. + \tag{26} +$$ +With one fixed $A>4$ and the corresponding fixed $B$ from section 6, +equation (26) proves (1) from Inputs S, M, H and P. -$$\sum_{p \le \sqrt{y}} \frac{g(p)}{p} -= \sum_{5 \le p \le \sqrt{y}} \frac{2}{p} -+ \sum_{z_0 < p \le z_1} \frac{2}{p} + O(1) -= 2\ln\ln y + 2(\ln\ln z_1 - \ln\ln z_0) + O(1). \tag{6.3}$$ +**Calibration of (1).** The proof of the reduction is displayed here. +Its analytic inputs are cited published statements. The specialization +is the existing project-derived, unrefereed claim [R2, R4, R6], not a +new theorem asserted to have passed external review. The implied +constant depends on the fixed choices made in this proof; we do not +claim a lower-bound constant uniform over arbitrary $A,B$. -The first term is $\Omega^{\mathrm{I}}$ charged over all $p \le \sqrt{y}$, with -$|\Omega^{\mathrm{I}}_p| = 2$ standing in for the source's $\ell_f$. The second -is $\Omega^{\mathrm{III}}$ over band 2, with $|\Omega^{\mathrm{III}}_p| = 2$ -standing in for its $M_p(f)$. The middle term of the source's three-way split, -the one carrying $h_f$, is identically zero. Both surviving terms are plain -Mertens sums over a constant, where the second was an application of Chebotarev -in the original. That is the whole of the simplification the substitution buys. +## 8. A shorter, weaker deduction -The $O(1)$ is an identified constant rather than a shrug. With Mertens' -constant $M = 0.2614972128476428$ and $\ln\ln\sqrt{y} = \ln\ln y - \ln 2$, +The same covering identity and sieve input also give +$$ + G_2(P(y))\gg y\log y. + \tag{27} +$$ +This is Theorem A of the live manuscript [R4, section 9]. +For completeness, let $m=\lfloor c y\log y\rfloor$ and +$z=\sqrt m$, with $c>0$ fixed and sufficiently small. +Choose $a_p=0$ for $p\leq z$. -$$\sum_{5 \le p \le \sqrt{y}} \frac{2}{p} - 2\ln\ln y \;\longrightarrow\; --2\ln 2 + 2M - 2\left(\tfrac12 + \tfrac13\right) = -2.529967,$$ +Lemma 2 with $\kappa=2$, together with Mertens' theorem, bounds the +uncovered indices by +$$ + C\,\frac{m}{(\log z)^2} + \leq (4Cc+o(1))\,\frac{y}{\log y}. + \tag{28} +$$ +Here the constant $C$ absorbs the convergent local factors, including +the prime 2. Since $z=o(y)$, the number of unused primes is +$\pi(y)-\pi(z)\sim y/\log y$. Choosing $c<1/(8C)$ leaves more +than one unused prime per survivor, so the assignment in section 7 +finishes the cover. -reached to $1.9 \times 10^{-5}$ at $\sqrt{y} = 10^7$ and to $0.0079$ at -$\sqrt{y} = 10^3$. At the $y$ where no computer can look, an explicit Mertens -bound does the same work. Rosser and Schoenfeld's Theorem 5 gives -$\ln\ln x + M - 1/(2\ln^2 x) < \sum_{p \le x} 1/p$ for $x > 1$ and -$\sum_{p \le x} 1/p < \ln\ln x + M + 1/(2\ln^2 x)$ for $x \ge 286$ ((3.17), -(3.18), read at the page image 2026-09-08); Dusart's Theorem 6.10 -(arXiv:1002.0442) sharpens the error to $1/(10\ln^2 x) + 4/(15 \ln^3 x)$, the -upper bound for $x \ge 10372$. $M$ cancels in a band difference, leaving, at -$A = 5$, a bound that falls from $2.475 \times 10^{-4}$ at $\ln y = 10^3$ to -$1.911 \times 10^{-5}$ at $\ln y = 10^9$ in Dusart's form, and from -$1.158 \times 10^{-3}$ to $9.314 \times 10^{-5}$ in Rosser and Schoenfeld's. +This argument does not use the smooth-number input, but it still uses +Input S. It cannot be described as having escaped the unread +Halberstam--Richert source dependency. Its composition has the same +project-derived, unrefereed status. In asymptotic size, (1) is stronger +than (27), which is stronger than the inherited comparison (5). -### 6.4 The assembly is an identity +## 9. Scope, corrections and remaining obligations -Write $\ell\ell = \ln\ln y$, $\ell\ell\ell = \ln\ln\ln y$, $\ell\ell\ell\ell = -\ln\ln\ln\ln y$. Then +A lower-bound construction does not bound $G_2$ from above. In +particular, (1) is $y^{1+o(1)}=o(y^2)$, but that fact does **not** +show $G_2(P(y))=o(y^2)$. It also gives no claim about a prescribed +prime-sized interval containing a twin pair. No assertion in this note +changes the project's standing upper-bound problem. -$$\ln\ln z_0 = \ln A + \ell\ell\ell, \qquad -\ln\ln z_1 = \ell\ell\ell\ell + \ell\ell - \ell\ell\ell - \ln A,$$ +This draft uses the live source record rather than reproducing the +historical quick draft unchanged: -so (6.3) reads $\sum g(p)/p = 4\ell\ell - 4\ell\ell\ell + 2\ell\ell\ell\ell - -4\ln A + O(1)$, hence +| Item | Treatment in this manuscript | +|---|---| +| Finite cofactor inequality | Retains $2(m+2)/y\leq z_0$, including the additive 2 | +| Mertens product | Uses $\asymp$, retaining the unspecified constant factors | +| Parameter constants | Fixes $A$, then $B$; does not claim uniformity over arbitrary choices | +| Printed corollary encoding | Replaced here by the counting weights (12), with Input S explicit | +| Selberg remainder discussion | Not an input to the quoted Brun-form estimate; no claim that dimension 4 is a unique support threshold | +| Smooth-number source | Uses [HT, Corollary 1.3] with its growing-$u$ range checked | +| Numeric threshold | No certified $y_0$ is given; the historical $10^{134.1}$ calculation with constants set to 1 is not such a certificate | +| Computation | No published numerical experiment or ladder has been rerun, and no asymptotic claim is called computationally verified | -$$\exp\!\left(-\sum g(p)/p\right) = \frac{A^4 (\ell\ell)^4}{(\ln y)^4 -(\ell\ell\ell)^2}, \qquad -S(m, \Omega) \ll m \prod_{p \le \sqrt{y}}\!\left(1 - \frac{g(p)}{p}\right) -\ll \frac{A^4 y}{B \ln y},$$ +The source record [R6] already explains why an earlier misattribution of +an explicit Mertens error does not change this asymptotic deduction. +Here no explicit Rosser--Schoenfeld or Dusart error formula is needed. +Likewise, the earlier finite tests of the covering step remain cited +historical evidence, not executions by this manuscript's preparer. -which is the source's own displayed $A^{2M(f) - 2h_f} y/(B\ln y)$ at -$M(f) = 2$, $h_f = 0$. Every coefficient cancels separately: in $\ln y$, in -$\ell\ell\ell$, in $\ell\ell$, and in $\ln A$. Since the implied constant -depends on $\kappa$ alone and not on $A$ or $B$, taking $B$ large against $A$ -gives $S(m, \Omega) \le y/(4 \ln y)$; with the ledger constant -$-2\ln 2 + 2M - 1/2 = -1.3633$ and the implied constant set to 1 this asks -$B \ge 4 e^{1.3633} A^4$, about $4.2 \times 10^3$ at $A = 4.05$, and §8 shows -that no tabulated threshold depends on $B$ in that range. +The next mathematical review should check Input S against a printed +upper-sieve theorem, including its uniformity for these changing +residue systems, and then inspect the reduction in sections 4--7. +Tracking all constants would be a separate task. Optimizing the bands +or proving an optimal growth law is also left open. No additional +logarithmic factor is asserted on the basis of a heuristic multi-cover +argument. -The collapse was checked numerically over 30 parameter triples out to -$\ln y = 10^{300}$, and again at machine precision by a second, independently -written pass: $\max |\text{ratio} - 1| = 5.684 \times 10^{-13}$. +## 10. Source and search record -### 6.5 The smooth-number step, and where $A > 4$ comes from +The following claim map separates proved elementary steps, imported +theorems and the unrefereed composition. -This is the one imported estimate. With $u = \ln m / \ln z_1 \sim A \ln\ln y / -\ln\ln\ln y$, the leading term of $u \ln u$ is $A \ln\ln y$, the -$\ln\ln\ln y$ cancelling, so $\Psi(m, z_1) = m (\ln y)^{-A + o(1)}$, the -$o(1)$ of order $\ln\ln\ln\ln y/\ln\ln\ln y$, and the -requirement $2\Psi(m+2, z_1) = o(y/\ln y)$ becomes - -$$(\ln y)^{3-A} \frac{(\ln\ln\ln y)^2}{(\ln\ln y)^4} = o\!\left(\frac{1}{\ln -y}\right), \qquad\text{that is}\qquad A > 4.$$ - -That is exactly the $\ell_f + M(f) = 4$ that the source's own version of this -step asks for. The estimate consumed is Hildebrand and Tenenbaum's Corollary -1.3 (§12), whose range hypothesis $u \le z_1^{1-\varepsilon}$ is slack -($u = 13.7$ against $z_1 = e^{23.2}$ at $y_0$); the tighter range of -Hildebrand's Theorem 1.1, measured by $\ln z_1 / (\ln\ln m)^{5/3}$, runs from -$6.364 \times 10^{-1}$ at $\ln y = 10^2$ to $2.339 \times 10^5$ at -$\ln y = 10^9$ and is not required. Bisecting for the -least admissible $A$ at each scale gives values rising through -$2.3196 \dots 3.8185$, approaching 4 from below, which is the expected -signature of an asymptotic condition met with room to spare only past the -threshold of §8. - -Combining, $R \le y/(4\ln y) + o(y/\ln y) \le y/(3\ln y)$. - ---- - -## 7. Band 3, and the end of the proof of Theorem B - -Each remaining $i$ is assigned its own prime $p \in (y/2, y]$ and the choice -$a_p \equiv i \pmod p$, which deletes $i$ and nothing else in $[1, m]$ that -matters. The assignment is available as soon as there are enough such primes. -Rosser and Schoenfeld give - -$$\pi(y) - \pi(y/2) \;>\; \frac{y}{\ln y} - \frac{0.62753\, y}{\ln y - \ln 2},$$ - -and the right-hand side exceeds $y/(3\ln y)$ exactly when -$0.62753 \ln y/(\ln y - \ln 2) < 2/3$, that is $\ln y > 11.807294$, that is -$y > 1.3423 \times 10^5$. That is far below any threshold of §8, so band 3 -never binds. - -Every $i \le m$ is now deleted by some prime $p \le y$. By Proposition 1, -$G_2(P(y)) \ge m + 1$, and $m$ is $y (\ln y)^3 (\ln\ln\ln y)^2/(\ln\ln y)^4$ -divided by the constant $B$. This proves Theorem B. $\square$ - ---- - -## 8. The threshold $y \ge 10^{134.1}$, and what it is a threshold for - -Every hypothesis used above is an explicit inequality in $L = \ln y$: - -| condition | inequality | where it is used | +| Claim | Mathematical status | Traceable source | |---|---|---| -| H1, $z_0 > 3$ | $A \ln L > \ln 3$ | §6.1, the $g(p) < p$ discharge | -| H2, $z_0 < z_1$ | $A^2 \ln^2 L < L \ln\ln L$ | §4, band 2 non-empty | -| H3, $z_1 < \sqrt{y}$ | $2\ln\ln L < A \ln L$ | §6.3, band 2 lies inside the sieve range $p \le \sqrt{y}$ of (6.3) and of Proposition 3(c) | -| H4, Case 1 | $\ln 2 - \ln B + 3\ln L + 2\ln\ell\ell\ell - 4\ln\ell\ell < A \ln L$ | (5.1), as coded with $2m/y$; the $2(m+2)/y$ form adds $\ln(1+2/m) < 10^{-100}$ | -| H5, smooth count | $\ln 2 + \ln(m/y) + \ln\rho(u) + \ln 12 + \ln L < 0$, the smooth term's share of the budget $y/(4\ln y) + 2\Psi + O(\sqrt y) \le y/(3\ln y)$; $\rho$ is de Bruijn's asymptotic, not an upper bound; slack by more than 17 in the logarithm at $L = 308.67$ (coded with $\ln 3$ until 2026-09-07; no tabulated value moved) | §6.5 | -| H6, band 3 | $L > 11.8073$ | §7 | -| H7, $z \ll X$ as $\sqrt{y} \le m$ | $\ln B + 4\ln\ell\ell - 3\ln L - 2\ln\ell\ell\ell < L/2$ | §6.1, the $z \ll X$ discharge; slack by 163 at $L = 308.67$, binds only for $\ln B > 163$; not coded in the bisection, which it cannot move | +| Covering identity and $G_2\geq g$ | Elementary proofs, section 2 | [R2, section 1]; [R4, section 2] | +| Bound (5) | Published one-class input plus elementary comparison | [KK, Theorem A and reference 1]; [R2, section 3] | +| Residue upper-sieve bound (11) | Elementary reduction conditional on the quoted Input S | [KK, Lemma 1 and Corollary 1, p. 4]; weights (12) | +| Three-band reduction and bound (1) | DERIVED / INFERRED, not refereed | [R2, section 4c]; [R4, sections 4--7]; [R6, section 3] | +| Smooth-number estimate and range | Published input, primary page inspected | [HT, Theorem 1.2 and Corollary 1.3, p. 417] | +| Weaker bound (27) | Derived from Input S, Mertens and PNT; not refereed | [R4, section 9] | +| Historical finite tests and source corrections | Reports of earlier project work only | [R4, sections 5 and 11]; [R6] | +| Absence of a twin-prime conclusion | Scope of the displayed argument | [R1]; sections 1 and 9 here | -Bisecting for the first $L$ at which all seven hold simultaneously, with every -implied constant set to 1: +**Search update, 18 September 2026.** We reused +`research/SEARCH-CONVENTIONS.md`, sections 1, 3 and 5, and the relevant +source-review and closed-route entries of `research/OUTCOMES.md`. +The additional web search used the author names and identifier +`2302.00459`, and the phrases "paired or two-class Jacobsthal", +"forbidden residues 0 and -2", and "OEIS A144311". We inspected the +arXiv v2 source paper, the journal's Math-Net text, the current OEIS +entry, and the primary smooth-number page identified below. The +Ziller--Morack abstract was reached as an adjacent object; its full +paper was not re-audited in this assignment. -| $A$ | $L_0 = \ln y_0$ | $y_0$ | binding condition | -|---|---|---|---| -| 4.05 | $3.0867 \times 10^2$ | $10^{134.1}$ | H2 | -| 4.5 | $4.0795 \times 10^2$ | $10^{177.2}$ | H2 | -| 5 | $5.3747 \times 10^2$ | $10^{233.4}$ | H2 | -| 7 | $1.2733 \times 10^3$ | $10^{553.0}$ | H2 | -| 10 | $3.1008 \times 10^3$ | $10^{1346.7}$ | H2 | +The search service incorrectly described [KK] as not journal-published. +The journal's own record and text resolve that point: it appeared in +*Izvestiya: Mathematics* in 2024. Search summaries are not used as +authority for the theorem or its bibliography. The inspected sources +do not themselves state this fixed-offset specialization as the +theorem under review. This bounded source comparison neither proves +novelty nor establishes the absence of other literature. -$B$ does not move $y_0$ to five figures, because it enters only H4, H5 and -H7, all of which are slack; the table is the same for every $B$ between 1 and -$10^{70}$, and the runs use $B = 10$. The binding condition is H2, $z_0 < z_1$, at every $A$, and H2 is a -property of the source's band geometry with nothing to do with the -substitution: the original construction has the same $y_0$ at the same $A$. +The 1974 Halberstam--Richert pages were not accessed in this assignment. +We retain the source registry's documented access limitation rather +than repeating its failed retrievals or presenting OCR as a page reading. +No new citation-count claim is made. -Two consequences, and both are limitations. +### Primary references -**$10^{134.1}$ is a floor on the true explicit threshold, not the threshold.** -It is what a reader gets by setting every implied constant to 1: the -fundamental lemma's constant at $\kappa = 4$, the smooth-number constant, the -$O(\sqrt{y})$ of (6.1), and the $O(1)$ of (6.3). A genuine explicit $y_0$ is -larger by an amount this project has not computed, so Theorem B as stated is -ineffective. +- **[KK]** A. B. Kalmynin and S. V. Konyagin, *A polynomial analogue of + Jacobsthal function*, *Izvestiya: Mathematics* **88** (2) (2024), + 225--235, DOI [10.4213/im9467e](https://doi.org/10.4213/im9467e). + [arXiv:2302.00459v2](https://arxiv.org/abs/2302.00459v2), + section 2, especially Lemma 1 and Corollary 1 on p. 4 and the + three-band construction on pp. 5--7. + [Journal text](https://www.mathnet.ru/eng/im9467). + Retrieved arXiv PDF SHA-256: + `9dd8ce68421e50756ff7341a2320aaf65f240a9d98cca127d0525113e1613a79`. +- **[HT]** A. Hildebrand and G. Tenenbaum, *Integers without large prime + factors*, *Journal de Theorie des Nombres de Bordeaux* **5** (1993), + 411--484, Theorem 1.2 and Corollary 1.3, printed p. 417. + [Numdam](https://www.numdam.org/item/JTNB_1993__5_2_411_0/). + Retrieved PDF SHA-256: + `f7641a11188d783d8e941883467d541d71429b98d6760e7d20bf85cf53e80bac`. + The displayed formula and range were read from that page image, + because text extraction omitted the mathematical displays. +- **[HR]** H. Halberstam and H.-E. Richert, *Sieve Methods*, Academic + Press, 1974, Theorem 2.2, pp. 68--69 as identified in the project + source record. Cited through [KK]. Original printed pages not read + in this assignment. +- **[O]** [OEIS A144311](https://oeis.org/A144311), definition and + attribution, revision 23 dated 5 December 2024, inspected + 18 September 2026. The current entry contains 22 values. They were + not recomputed or reproduced as a new dataset here. -**Band 2 is empty at every level any computer here can reach.** At $y = 4001$ -and $A = 4.05$ the parameters are $z_1 = 2.07$ against $z_0 = 5.26 \times -10^3$, so the interval $(z_0, z_1]$ is empty and the construction degenerates. -No finite computation in this project can exhibit it working as designed. In -particular the factor 4.0 to 6.6 by which a three-band certificate loses to a -plain greedy search at $y = 4001$ -(`research/history/staging/attack-lower-bound.md` D4) is not evidence about -Theorem B in either direction, and it must not be cited as any. +### Project sources and attribution -**The bound is a floor on the method rather than its ceiling.** Run the -source's Theorem 1 at $f(x) = x$, where $j_f$ is the ordinary Jacobsthal -function and $\ell_f = 1$, $h_f = 0$, $M(f) = 1$. It gives -$g(P(y)) \gg y \ln y \ln\ln\ln y / (\ln\ln y)^2$, weaker by exactly one factor -of $\ln\ln y$ than the one-class lower bound quoted in the source's own -introduction at p. 2, with the ratio measured here running from $4.6052$ at -$\ln y = 10^2$ to $20.7233$ at $\ln y = 10^9$. The slack comes from a fixed $A$ -in $z_0 = (\ln y)^A$ rather than an optimised Erdős and Rankin band choice, and -the transferred bound inherits it. An optimised band choice, or the Pintz and -FGKMT refinements imported into the substituted frame, should recover one or -two factors of $\ln\ln y$. That has not been attempted and is not claimed. +These are the public editions served on 18 September 2026. References +identify the existing work from which the manuscript was prepared. ---- +- **[R1]** + [Proposal: a two-class Erdos--Rankin lower bound for G2](/projects/twin-primes/docs/paper/proposals/prop-kk-lower-bound.md), + sections 1--6, grade QUICK-DRAFT. + SHA-256 `6e87abcd4d72d22eee1612cbe1f78e65d335e3b2d00c42eb6145587459198670`. +- **[R2]** + [Two-class lower bounds](/projects/twin-primes/docs/research/two-class-lower-bounds.md), + sections 1, 3 and 4c. + SHA-256 `d4af9ec9c9923a51f1c8478d898d1de3a7bed4be566c55f4c63d3d1c856f2556`. +- **[R3]** + [Historical quick draft](/projects/twin-primes/docs/paper/proposals/draft-kk-lower-bound.md), + especially sections 3--5. Superseded finite inequality is not reused. + SHA-256 `d3de944c59a8c885c599e89b8ba9d3c23c027f0b587f6719c37e1cb14e6373c5`. +- **[R4]** + [A lower bound for the two-class Jacobsthal function](/projects/twin-primes/docs/paper/kk-lower-bound.md), + live manuscript, sections 2--9 and 11--12. + SHA-256 `7c375d9510a22b6fc2c6241eeffe51c1d92cb9daa17e4eba63220ed1c29c107f`. +- **[R5]** + [Search conventions](/projects/twin-primes/docs/research/SEARCH-CONVENTIONS.md), + sections 1, 3 and 5. + SHA-256 `6160114056b5978f277959993d05a7693cc3c333bbe1becc57c4e3ed03db4b25`. +- **[R6]** + [Source review of Theorem 2c](/projects/twin-primes/docs/research/history/reviews-0907/11-two-class-theorem2c-source-review.md), + 8 September 2026, sections 0--3 and its source table. + SHA-256 `700031d9d550bb4d5067ec4691507585e4a149ed3d162b0e520e47318003fb43`. + The corresponding entry in + [OUTCOMES](/projects/twin-primes/docs/research/OUTCOMES.md) retains the + unread 1974 source page and unpriced constants. That edition has + SHA-256 `78c5ea9f7f96767696cdc0fed267fe06311c0ff723ab4ed15ef0c431343eb9e7`. +- **[R7]** + [Paper-suite architecture and authorship](/projects/twin-primes/docs/paper/PAPERS.md), + authorship and AI disclosure section, SHA-256 + `bd01edba6c84ac8681249c4c8cdc5e4cd845a73291406a2c38faa1cd56aeafc8`; + [mathematical style](/projects/twin-primes/docs/paper/writing-style-math.md). -## 9. Theorem A: the same object from published statements only +## Methods and AI disclosure -Theorem B follows a published construction and proves each step here, -consuming one estimate without an artifact. Theorem A consumes every -ingredient at its printed statement, and the composition is elementary and -written out here in full. It is two logarithms weaker, and the -trade is strength for provenance. - -*Proof of Theorem A.* Let $m = c_0\, y \ln y$ with $c_0$ fixed at step 3 below, -and let $z = \sqrt{m}$. - -**1. The survivor count, from Corollary 1 consumed as a theorem.** Apply the -corollary quoted in §6.1 with $X = m$, $\Omega_2 = \{0\}$, and -$\Omega_p = \{0, -2 \bmod p\}$ for odd $p \le z$. Then $g(2) = 1 < 2$, -$g(p) = 2 < p$ for odd $p$ (the two classes being distinct for $p$ odd), -$\kappa = 2$, and $z = \sqrt{m} \ll m = X$, so every hypothesis of Lemma 1 is -discharged. The stage-1 survivors, that is the twin slots of $[1, m]$ with -respect to the primes $\le z$, number -$\#V \le C_1 \, m \, V(z)$ with $V(z) = \prod_{p \le z}(1 - g(p)/p)$. - -**2. The product, by Mertens.** Writing -$1 - 2/p = (1-1/p)^2 \cdot \big[(1-2/p)/(1-1/p)^2\big]$ with the bracket -convergent, Mertens' third theorem gives -$\prod_{2 < p \le z}(1 - 2/p) = (C_2 + o(1))/\ln^2 z$. At $z = \sqrt{m}$ this -carries a factor 4, so $\#V \le (4C_3 + o(1))\, m/\ln^2 m$. - -**3. The mop-up, one prime per survivor.** Since $\sqrt{m} = o(y/\ln y)$, the -prime number theorem gives $\pi(y) - \pi(\sqrt{m}) = (1 + o(1))\, y/\ln y$. -From step 2 and $\ln m \sim \ln y$, -$\#V \le 4 C_3 c_0 (1 + o(1))\, y/\ln y$. Choose $c_0 = 1/(8C_3)$. For large -$y$ there is then an injection $V \to \{\text{primes in } (\sqrt{m}, y]\}$, -$r \mapsto p_r$. Set $a_{p_r} = r \bmod p_r$, and $a_p = 0$ for every other -$p \le y$. - -**4. Every $r \in [1, m]$ is covered.** A non-survivor has some $p \le -\sqrt{m}$ with $p \mid r$ or $p \mid r+2$, that is $r \equiv a_p$ or -$r \equiv a_p - 2 \pmod p$ at $a_p = 0$. A survivor is covered by its own -prime. By Proposition 1 a full cover of $[1, m]$ by pairs $\{a_p, a_p - 2\}$ -over $p \le y$ exhibits a run of $m$ consecutive residues of $P(y)$ containing -no twin slot, so $G_2(P(y)) \ge m + 1 = (c + o(1))\, y \ln y$. $\square$ - -All four constants are effective, so $c$ is effective, which Theorem B's is -not. - -**Custody.** This chain was written on 2026-08-20, held out of every live -document until a dedicated adversarial pass reported, and confirmed by it at -every ingredient: the Corollary 1 statement re-read at a freshly fetched page -image with a matching md5, the Mertens and prime number theorem steps -re-derived, the hypothesis list discharged item by item, and the Chinese -remainder identity re-derived from its own proof -(`research/history/staging/redteam-0820-math.md` §3.3). The pre-registration -was committed alone before any producer or report existed, so the custody is -provable rather than declared. - -**The finite end-to-end check.** The whole pipeline was rebuilt from the -pre-registration text alone, by separately written code, and the assembled -cover replayed through an independent per-residue scan: $|V| = 9889$, -$\prod(1-2/q) = 0.1750499$, mean leftover $1730.83$ against an expectation of -$1731.07$, best trial 1654, a mop-up of 1153 primes at 1.43 slots per prime, -largest prime used $10861$, **uncovered points in $[1, 200000]$: zero**, and -ratio $m/(y \ln y) = 1.9816$. That is an end-to-end correctness check of the -construction at one scale. It is not evidence for the limit, and the record -says so in the same sentence -(`research/history/staging/redteam-0820-math.md` §3.4). - -Theorem A supersedes nothing in the record except its own earlier sketch: -`research/two-class-lower-bounds.md` §4b had this accounting at INFERRED -grade, reaching $m \asymp y \ln y$ by the same two-stage argument. What §9 adds -is the proof grade, the chain of custody, and the finite echo. - ---- - -## 10. What these bounds do not do - -**They say nothing about the upper side.** The best two-class upper bound -available is $G_2(P(y)) \ll_\varepsilon y^{\beta_2 + \varepsilon}$ at -$\beta_2 = 4.26645028414864191641$, the sifting limit of the dimension-2 -Diamond, Halberstam and Richert sieve. Nothing in §§4 to 9 constrains it, in -either direction, because a covering construction is a lower-bound device and -the sifting limit is a positivity threshold for a lower-bound sieve. The two -sides do not meet: - -| | statement | exponent of $y$ | -|---|---|---| -| upper | $G_2(P(y)) \ll_\varepsilon y^{4.26645 + \varepsilon}$ | $4.266\ldots$ | -| lower, Theorem B | $G_2(P(y)) \gg y (\ln y)^3 (\ln\ln\ln y)^2/(\ln\ln y)^4$ | $1 + o(1)$ | - -so the gap is a factor $y^{3.266 + o(1)}$, and moving Theorem B by another -logarithm, or by ten, does not narrow it in the exponent. On the evidence -available the truth is nearer the lower side: the exponent measured on the 22 -trusted terms of A144311 is $1.50 \pm 0.05$ in the $y$-frame, and the estimator -that produced it is provably biased upward by about $+0.28$ on a control object -whose answer is 1 (`research/exponent-control.md` §§1, 5). A measured exponent -with a known upward bias of that size is not an argument about the limit, and -it is recorded here only so that the reader knows which end of the interval the -data sits at. - -**They say nothing about the twin prime conjecture.** No lower bound on -$G_2$ can, since a lower bound says the adversary can build long twin-free -runs, which is the direction away from producing twin primes. What a *bound* on -$G_2$ would decide is the internal Zone Postulate of this project, which needs -$G_2(P(y)) < y'^2$ where $y'$ is the next prime; the theorems above leave that -margin exactly where it was, since $y^2/(y\ln^3 y) \to \infty$ and the margin -is still $y^{1-o(1)}$. A stronger construction-side bound was the one thing -that could have threatened the Postulate. Theorem B is not it. - -**They are not effective.** Theorem A's constant is effective; its threshold -is not stated. Theorem B's $y_0 = 10^{134.1}$ is a floor with every implied -constant set to 1 (§8), and no effective version exists. - -**They are not exhibited.** No computation in this project reaches band 2 -(§8), so the only empirical support for Theorem B is the brute force of -Proposition 3 rather than of the construction, and a referee is entitled to -call the theorem unfalsifiable by anything in this repository. That is -accurate. Theorem A is different: its construction runs at accessible scale and -was replayed clean (§9), which is the honest reason to prefer it despite the -two logarithms. - -**They do not improve the best constructed value.** The certified ladder in -this project reaches $356{,}712$ at $y = 4001$, replayed by an independent -routine with zero uncovered points at all sixteen levels -(`research/G2-STATE.md` §3a), and a further certified $479{,}339$ at -$y = 5003$ (`research/two-class-lower-bounds.md` §0). Those are finite -constructions, and neither theorem above predicts them or is predicted by them. - ---- - -## 11. Residuals, and what would break each theorem - -### 11.1 The two load-bearing readings of the source (Theorem B) - -Theorem B rests on two sentences of someone else's paper: that Corollary 1's -conclusion depends on $\Omega_p$ only through $g(p) = |\Omega_p|$, and that -$\Omega^{\mathrm{II}}_p$ is defined by the non-linear irreducible factors of -$f$, hence empty for $f = x(x+2)$. Both were read from 200 dpi page images of -the PDF of record by the first adversarial pass, and the first of them was -re-read a third time by the pass that confirmed Theorem A. The second was not -re-read by that pass, which had the PDF only for the Corollary 1 page. If -either reading is wrong the substitution does not go through, and no -computation available here would show it. - -*Re-read at source 2026-09-07* (arXiv:2302.00459v2 PDF, md5 -`b5d7d2a23ffd902415057adebfe430b1`, matching the artifact recorded in -`research/history/staging/lit-pdf-kalmynin-konyagin.md`; full text, pp. 1 to -7 line by line). Both readings are CONFIRMED. Corollary 1 as printed -imports only Lemma 1's hypotheses, $g$ multiplicative with $g(p) \le \kappa$ -and $g(p) < p$, $|r_d| \le g(d)$ for $d \mid P(z)$, $z \ll X$, together with -"for any $p \le z$ the set $\Omega_p \subset \mathbb{Z}/p\mathbb{Z}$ contains -$g(p)$ elements". The polynomial $f$ does not appear in Lemma 1, Corollary 1 -or its four-sentence proof; it enters only the proof of Theorem 1 (the -trichotomy on p. 6 and the evaluation of $\sum g(p)/p$ on pp. 6 to 7), -which this manuscript replaces by its own §3 to §5. $\Omega^{\mathrm{II}}_p$ -is defined on p. 5 as the residues at which "some non-linear irreducible -factor $q(x)$ of $f(x)$" vanishes, empty for $x(x+2)$. The published -Izvestiya English edition (88:2, 225 to 235; mathnet full text, md5 -`9e7f3c54b1979cdfb505c14b4576c4e0`, which now needs a browser User-Agent -to fetch) was read at extracted text for the Corollary 1 material on -2026-09-07 and carries the same statement and proof text. - -### 11.2 Halberstam and Richert Theorem 2.2, unread (Theorem B) - -Nobody in this project has read it at source, and the record says so rather -than implying otherwise: `research/PRIOR-ART.md` records the companion -Corollary 2.4.1 as UNREACHABLE after fourteen access routes, corroborated by -Ford's course notes rather than verified. Until 2026-09-07 this section also -carried a Selberg-form support correction (§6.2 as it then stood) and the -claim that $\kappa = 4$ was load-bearing with a margin one sieve dimension -wide. Both were about a theorem the source does not cite; the paragraph below -records what the source re-read found, and §6.2 is now a remark. The residual -that remains is only the unread 1974 page behind a printed, refereed Lemma 1, -and it is shared with Theorem A, whose step 1 consumes the same statement at -$\kappa = 2$. - -One more residual on the source, found by the red team of 2026-09-07 and -recorded here because it lands on both theorems equally: the printed proof of -Corollary 1 encodes avoidance of $\Omega_p$ by the product $\prod_{p \le z} -\prod_{r \in \Omega_p}(P(z;p)\,n + r\,Q(z;p))$, and the representatives $r$ -are left as elements of $\mathbb{Z}/p\mathbb{Z}$; read as integers in -$[0, p-1]$, any $2 \le r \le p-1$ has a prime factor $q 0$, $S(\mathcal A, \mathfrak p, z) \ll X \prod_{p 1$, consumed in §7; Theorem 5, (3.17), - (3.18), the explicit Mertens bound with error $1/(2\ln^2 x)$, the upper - bound for $x \ge 286$, consumed in §6.3.* -- P. Dusart, *Estimates of some functions over primes without R.H.*, - arXiv:1002.0442, Theorem 6.10: $|\sum_{p \le x} 1/p - \ln\ln x - B| \le - 1/(10\ln^2 x) + 4/(15\ln^3 x)$, the upper bound for $x \ge 10372$. *Read at - extracted text 2026-09-08; this is the form §6.3's figures use.* -- A. Hildebrand and G. Tenenbaum, *Integers without large prime factors*, - J. Théorie Nombres Bordeaux **5** (1993) 411–484, numdam - `JTNB_1993__5_2_411_0`. *Read at page images pp. 414–417 on 2026-09-08 (md5 - `a036c61d06a8199185379c883059566a`). Theorem 1.2: - $\log(\Psi(x,y)/x) = \{1 + O(\exp(-(\log u)^{3/5-\varepsilon}))\}\log\rho(u)$ - uniformly for $y \ge 2$, $1 \le u \le y^{1-\varepsilon}$, the upper bound - implicit in de Bruijn (1966); Corollary 1.3: $\Psi(x,y) = x\,u^{-(1+o(1))u}$ - as $y, u \to \infty$ uniformly for $u \le y^{1-\varepsilon}$. §6.5 consumes - Corollary 1.3 at $x = m + 2$, $y = z_1$, where - $u \sim A\ln\ln y/\ln\ln\ln y$ and the range hypothesis is slack. Theorem - 1.1 (Hildebrand 1986) with its range $u \le \exp((\log y)^{3/5-\varepsilon})$ - is not needed.* - -**The object and its data.** - -- OEIS **A144311** (A. Carter, September 2008, $a(1)$ to $a(7)$; - M. Alekseyev, 2009, $a(8)$ to $a(16)$; Wang, 2024, $a(17)$ to $a(22)$), 22 - terms to $y = 79$, equal to $G_2(P(y)) - 1$. *Entry read; it carries no - formula line and no reference line.* -- OEIS **A048670** (Jacobsthal at primorials), **A288815** and **A072753** - (paired and free two-class), **A059861** (twin-slot census). -- L. Ziller and J. Morack, arXiv:1706.00317 and arXiv:1706.03668, with - `full_details.pdf`. *Read; they carry an upper-bound conjecture and no lower - bound.* -- T. R. Hagedorn, *Math. Comp.* **78** (2009), Proposition 1.1. *One class.* -- A. Kourbatov, arXiv:1301.2242. *The maximal twin-gap law, conjectural, and it - is $0.76\log^3 p$.* - -**The frame.** - -- F. B. Holt and H. Rudd, arXiv:1408.6002; F. B. Holt, arXiv:2603.25915; - primegaps.info; code and data at github.com/fbholt/Primegaps-v2 (73 MB, - cloned 2026-08-17). *The cycle of gaps $G(p\#)$, the fold recursion and its - closure theorem, the transfer operator with binomial eigenvectors, and the - interval of survival $[p_k^2, p_{k+1}^2]$. Priority is theirs; see - `research/PRIOR-ART.md`.* - -**Discussion.** - -- Erdős problems #687 (US\$1000, the order of $Y(x)$) and #970 (the order of - $h(k)$); MathOverflow 88323 (*Analogues of Jacobsthal's function*), the - question whose convention owns the bounded-classes-per-prime formulation. - *All re-checked at the pages 2026-08-18.* - -**Footnote on [MEMORY].** Two entries above are marked [MEMORY]: Rankin 1938 and Pintz 1997. Halberstam and Richert's *Sieve Methods* Theorem 2.2 is marked [OCR ONLY]: its bibliographic details are confirmed (OpenLibrary and Zbl 0298.10026) and its statement is reached at an OCR index, which is not a reading; the page image is the owed item (§11.2). Two -former [MEMORY] entries, the Rosser and Schoenfeld numbering and the -smooth-number estimate, were read at page images on 2026-09-08; the first was -misattributed (`research/history/reviews-0907/11`, F1). For each, -this project holds the statement and uses it, and holds no page image, PDF or -photograph confirming the bibliographic details as written. The distinction -matters here because this project has already retracted one constant that came -from an arXiv HTML rendering rather than from a PDF, and the rule adopted after -that (`research/PRIOR-ART.md`, "Provenance: what artifact was actually read") -is that a quotation records which artifact it was read from. These five record -that no artifact was read. - ---- - -## Appendix A. Provenance of every number - -Every figure printed above, in order of appearance, with the corpus file or -cited page it comes from. Producers named here carry embedded OUTPUT blocks -written by `research/qc/embed.js`, so each figure below the banner is bound to -the code above it by a `code-sha256`; no number in this note was pasted by -hand from a terminal. - -| number | where it appears | source | -|---|---|---| -| $\beta_2 = 4.26645028414864191641$ | §1.2, §3, §10 | Booker and Browning, via `research/PRIOR-ART.md`; the tract's p. 79 prints $4.266$ | -| $\beta_1 = 2$, Iwaniec's exponent | §1.1 | `research/covering-dive.md`, "where the proof breaks" item 1 | -| gap factor $y^{3.266+o(1)}$ | Abstract, §10 | arithmetic on the two rows of §10's table | -| $G_2/g$ = 2.00, 3.00, 3.00, 3.00, 4.15, 4.41, 5.10, 5.61, 6.00, 8.00, 7.38 at $y = 5 \dots 41$ | §2 | `research/two-class-lower-bounds.md` §1, VERIFIED at all eleven shared terms | -| A144311, 22 terms to $y = 79$; Carter 2008 $a(1..7)$, Alekseyev 2009 $a(8..16)$, Wang 2024 $a(17..22)$ | §1.2, §12 | `research/PRIOR-ART.md`; `research/SEARCH-CONVENTIONS.md` §§1, 2 | -| $\mathrm{Res}(x, x+2) = 2$; $\mathrm{Res}(x(x+2),(x-1)(x+1)) = -3$; $\mathrm{Res}(x(x+2),(x-a)(x-a+2)) = a^2(a^2-4)$ | §4 | `research/attack-kk-substitution.js`, §§C, D | -| 78,498 primes below $10^6$, zero exceptions to $\lvert \Omega^{\mathrm{III}}_p\rvert = 2$ | §4, §11.4 | `research/attack-kk-substitution.js` | -| overlap only at $p = 2, 3$, by enumeration to 1000 | §6.1, §11.4 | `research/attack-kk-substitution.js` | -| brute force at $y = 200000$, $z_1 = 300$, $m = 4\times10^6$, threshold $2m/y = 40$; unkilled 16985 / 14381 / 13164 / 11228 / 10372 / 9086 and counterexamples 0 / 0 / 0 / 235 / 799 / 2921 at $z_0 = 100, 60, 45, 30, 20, 10$ | §5 | `research/verify-kk-substitution.js`; report `research/history/staging/verify-kk-substitution.md` | -| $3\kappa - 1 = 11$ at $\kappa = 4$; $\xi = \sqrt{y}/(\ln y)^7$; $C > 3\kappa/2$ | §6.2 (remark) | second adversarial pass, `research/history/CHANGELOG.md` 2026-08-19, which had no source PDF and priced a Selberg-form remainder Lemma 1 does not carry; demoted 2026-09-07, `research/history/reviews-0907/05`, `06` | -| $M = 0.2614972128476428$ | §6.3 | Mertens' constant, standard; used by `research/attack-kk-substitution.js` | -| limit $-2\ln2 + 2M - 2(\frac12+\frac13) = -2.529967$; residual $0.0079$ at $\sqrt{y}=10^3$ falling to $1.9\times10^{-5}$ at $\sqrt{y}=10^7$ | §6.3, §11.4 | `research/attack-kk-substitution.js` | -| explicit Mertens error $1/(10\ln^2x) + 4/(15\ln^3x)$ (Dusart, Theorem 6.10, $x \ge 10372$; Rosser and Schoenfeld's (3.18) has $1/(2\ln^2 x)$, $x \ge 286$); evaluated at $A = 5$: $2.475\times10^{-4}$ at $\ln y = 10^3$ and $1.911\times10^{-5}$ at $\ln y = 10^9$ | §6.3 | statements per §12, read 2026-09-08; evaluations by `research/attack-kk-substitution.js` §F4 (the prose there still names Theorem 20) | -| exponent assembly, 30 triples to $\ln y = 10^{300}$, $\max\lvert \text{ratio}-1\rvert = 5.684\times10^{-13}$ | §6.4 | first pass plus an independent second pass at machine precision, `research/history/CHANGELOG.md` 2026-08-19 | -| Hildebrand slack $\ln z_1/(\ln\ln m)^{5/3}$: $6.364\times10^{-1}$ at $\ln y = 10^2$, $2.339\times10^5$ at $\ln y = 10^9$ | §6.5 | `research/attack-kk-substitution.js` | -| least admissible $A$ rising $2.3196 \dots 3.8185$ | §6.5 | `research/attack-kk-substitution.js`, bisection per scale | -| $0.62753$; $\ln y > 11.807294$; $y > 1.3423\times10^5$ | §7 | Rosser and Schoenfeld Corollary 1, (3.5) and (3.6), read at page image 2026-09-08, plus arithmetic in `research/attack-kk-substitution.js` | -| $y_0$ table: $L_0 = 3.0867\times10^2$, $4.0795\times10^2$, $5.3747\times10^2$, $1.2733\times10^3$, $3.1008\times10^3$ at $A = 4.05, 4.5, 5, 7, 10$; $y_0 = 10^{134.1}, 10^{177.2}, 10^{233.4}, 10^{553.0}, 10^{1346.7}$; H2 binding throughout | §8 | `research/attack-kk-substitution.js`, seven-inequality bisection, reproduced independently by the second pass | -| $z_1 = 2.07$ against $z_0 = 5.26\times10^3$ at $y = 4001$, $A = 4.05$ | §8 | live-layer corrected value, `research/two-class-lower-bounds.md` §4c. The frozen `attack-kk-substitution.md` §5 still carries the pre-correction $z_0$ figure, three orders larger, with no run behind it; the conclusion is unaffected because $z_0$ already exceeds $y$ there | -| factor 4.0 to 6.6, three-band certificate against plain greedy at $y = 4001$ | §8 | `research/history/staging/attack-lower-bound.md` D4, cited only to be excluded as evidence | -| ratio $4.6052$ at $\ln y = 10^2$ to $20.7233$ at $\ln y = 10^9$ for the method's own one-class shortfall | §8 | `research/attack-kk-substitution.js` | -| $c_0 = 1/(8C_3)$ | §9 | `research/history/staging/import-hypergraph.md` §4 step 3 | -| $\lvert V\rvert = 9889$; degree $1.708115$; $\prod(1-2/q) = 0.1750499$; $E = 1731.07$; mean leftover $1730.83$; trial variance ratio $0.43$; best trial 1654; mop-up 1153 primes at 1.43 slots per prime; largest prime $10861$; uncovered in $[1, 200000]$ zero; ratio $1.9816$ | §9 | `research/history/staging/redteam-0820-math.md` §3.4, producer `rt0820-t3-cover.js`, rebuilt from the pre-registration text alone | -| exponent $1.50 \pm 0.05$ on 22 trusted terms; control bias $+0.28$ | §10 | `research/exponent-control.md` §§1, 5 | -| certified $356{,}712$ at $y = 4001$, sixteen levels, zero uncovered | §10 | `research/G2-STATE.md` §3a | -| certified $479{,}339$ at $y = 5003$ | §10 | `research/two-class-lower-bounds.md` §0 | -| md5 `b5d7d2a23ffd902415057adebfe430b1`, 12 pages, 148,566 bytes | §12 | `research/history/staging/lit-pdf-kalmynin-konyagin.md` §0 | -| zero forward and reverse citations on three DOIs (OpenAlex); Semantic Scholar empty twice against calibration positives 46, 48, 5; A048670 carries five formula lines and one reference line | §11.3 | `paper/proposals/prop-kk-lower-bound.md` §4, run 2026-08-19 | - -**One attribution this note declines to make.** Section 8 compares the -method's own one-class output against "the one-class lower bound quoted in the -source's own introduction at p. 2". This project's records name that quoted -bound two different ways, Ford, Green, Konyagin and Tao in -`paper/proposals/prop-kk-lower-bound.md` §6 and Ford, Green, Konyagin, Maynard -and Tao in `paper/proposals/draft-kk-lower-bound.md` §6, and the source page -has not been re-read to settle which. The comparison is unaffected, since the -two published bounds agree in the shape being compared, and the name is left -unstated rather than guessed. - ---- - -## Appendix B. Custody residuals in the underlying records - -Three residuals sit in the frozen staging records rather than in the -mathematics, and a reader following the citations will meet them. This project's -document convention keeps frozen records frozen and corrects in the live layer, -so the disagreements below are by design and are listed rather than repaired. - -1. `research/history/staging/attack-kk-substitution.md` §5 carries a $z_0$ - figure at $y = 4001$ with no run behind it, three orders too large. The - corrected value is in `research/two-class-lower-bounds.md` §4c and is what - §8 above uses. -2. That file and the first adversary's report use opposite conventions for - $p_0$, each matching its own producer, and the two have not been reconciled. -3. The frozen first adversary's report reads "necessary and sufficient" for the - Case 1 inequality (5.1) where the live layer, and §5 above, read - "sufficient" only. The brute force supports the weaker reading and not the - stronger one. - -A draft quotes the live layer. This one does. - ---- - -## Appendix C. Registry updates this draft implies - -Listed rather than made, since this note is written under a fence that permits -no edits to existing files. - -- `paper/proposals/PROPOSALS.md`: the `prop-kk-lower-bound.md` row's grade stays - QUICK-DRAFT and its "last regraded" date stands, since a draft existed - already; the row's pointer should name `paper/kk-lower-bound.md` as the draft - of record beside `paper/proposals/draft-kk-lower-bound.md`, which becomes the - earlier, shorter version rather than the draft. -- `paper/proposals/PROPOSALS.md`: `prop-xlnx-lower-bound.md`'s first upgrade - trigger reads "upgrade to QUICK-DRAFT when the paper case is written". Section - 9 above writes it, inside a note that states both bounds with their grades, - which is what that trigger asked for. The grade should move PROPOSAL to - QUICK-DRAFT, with the regrade recorded in `research/history/CHANGELOG.md` per - the registry's regrade rule. -- `paper/PAPERS.md`: the suite has no entry for a two-class lower-bound note. - Either it becomes a fifth paper or it folds into Paper II, which already - carries the pointwise $G_2 \ge g$ bound and is the note that would then be - two-sided at three levels rather than two. That is a call for Chris, and the - reason to prefer folding is that a referee reading a lower-bound note wants - the upper bound in the same document. -- `research/two-class-lower-bounds.md` §4c and `research/G2-STATE.md` §3a: both - carry the two theorems already, at the grades used here; the only change is a - pointer to this draft. - ---- - -*This document states current understanding. Superseded claims, retired numbers -and the reasons they changed are in -[research/history/CHANGELOG.md](../research/history/CHANGELOG.md), indexed by -document.* +The framework, vocabulary and driving questions are the author's, +developed over six years of independent work, as recorded in [R7]. +Formal derivations, literature audits, computations and manuscript +drafting in the project have used AI assistance under the author's +direction. This revision was prepared with GitHub Copilot CLI using +GPT-6 Astra. Its mathematical work consisted of source inspection, +exposition and checking the displayed deductions, not a new numerical +experiment. Earlier computations have their own cited code and +recorded outputs. Finite checks do not establish the asymptotic +argument. Refuted intermediate steps and the remaining source-access +and referee obligations are retained explicitly.