---

Subject and independence

Role. Independent reviewer in a fresh context, given only the review assignment; took no part in writing the reviewed page or any page in its folder, and had not seen the source before this review. Charge: refutation. People are identified by role only.

Frozen subject. wiki/research/erdos_49/theorem_1_2_reconstruction.md as it stood on 2026-09-28T05:03:27Z, read from the committed text (the worktree's checked-out state was the same one). The page is the Theorem 1.2 reconstruction.

Artifact. The 17-page author manuscript PDF held by the library card Pollack, Pomerance and Treviño (2013), "Ramanujan Journal manuscript No. (will be inserted by the editor)", whose printed page numbers coincide with the physical pages. The PDF under the second card of the same paper, the arithmetic-functions card, is byte-identical (same size and digest), so the locators apply to both. Physical pages 1--12 were read on the text layer (pdftotext -layout). Page images of all 17 pages were rendered at 130 dpi; the images of pages 1, 2, 5, 6, 7, 8, 9 and 10 were read clause by clause for every displayed formula the page cites: the definition of $M^\uparrow(x)$ (p. 1); Theorem 1.2 and the paragraph on §3 and on Erdős's count (p. 2); the definition of $P(x;k)$, Theorem A, the split into $P_0$ and $P_1$, Theorem B and $c(k)$ (p. 5); Theorem C, Theorem 3.1 with its sketch, Lemma 3.2, Theorem 3.3 and Remark 3.1 (p. 6); the proof of Theorem 3.3, the footnote on Ford's corrected arXiv version and Ford's $Z(x)$ (p. 7); the constants $C$, $C'$ and Lemma 4.1 (p. 8); Lemma 5.1 with its proof (p. 9); the proof of Theorem 1.2 (p. 10). Bibliography entries [4], [8] and [10] were read on the text layer of pp. 16--17 to check the page's two external citations.

Allowed material actually read. In the same state, the three input pages the page cites: Lemma 5.1 (Statement, and its proof because input (A) is consumed whole and its uniformity in $S$ is load-bearing), Theorem 3.1 (Statement) and Theorem 3.3 (Statement, Definitions, and Steps 0 and 0' because the deduction of input (B) uses the absolute bound $c(k)\le c^*$ derived there). The provenance paragraphs of the two cards named above and the Statement section of the primes card's Theorem 1.2 page. The statement paragraph of Problem 49. docs/verification.md "Whole-claim report" and "Audit checklist" (the Erdos-specific subsections and the shared list of canonical failure modes), docs/evidence.md "Source fidelity", and docs/math_authoring.md in full. For the mechanics check of one wikilink whose label breaks across a line, the wiki tool's link-parsing regular expression was inspected and the corpus in the frozen state was pattern-counted (355 pages carry the same shape); no page content was read for that count.

Exposures. Four, all disclosed and none affecting the mathematical verdict. (1) The two card _index.md files and the Theorem 1.2 result page were printed in full, so text beyond their provenance paragraphs and Statement sections reached this reviewer: the primes card's "Read status" paragraph, its Contents list and its "Relation to E49" section; the result page's "Source", "Read depth", "Proof pointer", "Dependencies" and "Bears on" sections; and the arithmetic-functions card's whole body, including its one-line "Living verification" sentence. (2) The Problem 49 page has no Statement heading; the span above its "Current assessment" heading holds a "Status" paragraph, the Source and References entries and a Formalization paragraph, all of which reached this reviewer with the statement. (3) The three input reconstruction pages were printed in full, so their Standing paragraphs (each "author-recorded reconstruction; not an independent review") were seen. (4) Two other review files exist in this directory; they were listed by name only and not opened. Nothing under any evidence/ folder was read, no Current assessment or Known results section was read, and no web search was made.

Restatement

Let $\varphi$ be Euler's totient. For real $x\ge1$ let $\mathcal W(x)$ be the set of values $\varphi(n)$ over the positive integers $n\le x$, and $W(x)=\#\mathcal W(x)$. Call a set $S$ of positive integers nondecreasing when $\varphi(m)\le\varphi(m')$ for every pair $m<m'$ in $S$ (ties allowed; the empty set and singletons qualify), and let $M^\uparrow(x)$ be the largest cardinality of a nondecreasing $S\subseteq\{1,\ldots,\lfloor x\rfloor\}$, a maximum over finitely many sets. The theorem (source p. 2) asserts

$$\limsup_{x\to\infty}\frac{M^\uparrow(x)}{W(x)}<1 .$$

As reconstructed on the page, the quantitative content is: there is an absolute constant $c$ with $0<c\le1$ (the constant of Lemma 5.1), an absolute constant $B$ and an absolute threshold $X$ such that for every $x\ge X$ and every nondecreasing $S\subseteq[1,x]$,

$$\#S\le1+\frac{(1+B)x}{\log x}+(1-c)W(x),$$

and $x/\log x=o(W(x))$; hence $M^\uparrow(x)\le(1-c+o(1))W(x)$ with an $o(1)$ that does not depend on $S$, and for each fixed $c'<c$ the bound $M^\uparrow(x)\le(1-c')W(x)$ holds for all large $x$. Conventions: $\log_k$ is the $k$-th iterated natural logarithm; implied constants are absolute (source p. 3); a "totient" is a value of $\varphi$; $P(x;k)$ counts the positive integers $n\le x$ with $\varphi(n)=\varphi(n+k)$. The page's consequence, using Erdős's count $W(x)=x/(\log x)^{1+o(1)}$ as an imported theorem, is $M^\uparrow(x)=o(x)$ and, with $M^\uparrow(x)\ge\pi(x)$, $M^\uparrow(x)=x/(\log x)^{1+o(1)}$.

Checklist

• Quantifiers and scope. Pass. Every "for large $x$" of the source is turned into a named threshold ($x_0$ for (A), $x_1$ for (B), their maximum for the main argument), the uniformity "in the choice of $S$" of Lemma 5.1 becomes absolute constants, the range $k\le\log x$ is respected in every use of (B), and the limit superior is derived from an eventual bound rather than asserted. One boundary case ($\#S\le1$) is silently assumed away by "$m=\#S\ge2$"; it is trivial and is filed as note F3.
• Circularity. Pass. The argument consumes Lemma 5.1 (missing totients), the collision bound (B) and Ford's order of magnitude; none is equivalent to the theorem, and no step assumes a bound on $M^\uparrow(x)$.
• Model and convention changes. Pass. "Nondecreasing" is weak monotonicity on the page and in the source (p. 1). The Problem 49 section transports from strict to weak sets by inclusion (a strictly increasing set is nondecreasing), which is the correct direction for an upper bound. The counting of index pairs by $P(x;k)$ uses the source's own definition of $P(x;k)$ (p. 5), and $P_0$, $P_1$ are the source's.
• Finite and statistical overreach. Inapplicable. No finite verification, computation or heuristic enters the page; the explicit totient pair on the Lemma 5.1 page is consumed only through the interface (A).
• Uniformity. Pass. (A) carries absolute $c$ and $x_0$; (B) carries absolute $B$ and $x_1$ because the function $\varepsilon(x)=(\log x)^{-1/2}$ is fixed before Theorem 3.3 is invoked and $c(k)\le c^*$ is absolute (Remark 3.1, derived on the Theorem 3.3 page); the $o(1)$ in the conclusion is $(1+(1+B)x/\log x)/W(x)$, which does not depend on $S$, so taking the maximum over $S$ is legitimate. The dependence of each constant is stated.
• Extremal conclusions. Pass. $M^\uparrow(x)$ is a maximum over finitely many subsets, so it exists; the page claims no sharpness and no value of $c$.
• Consequences and composition. Pass with two suggested corrections. The "hence" sentences were attacked one by one: $M^\uparrow(x)=o(x)$ follows from $M^\uparrow(x)\le W(x)$ for large $x$ and (D); $M^\uparrow(x)=x/(\log x)^{1+o(1)}$ follows with $\pi(x)$ below; clause (iii) of Problem 49 follows by inclusion; the negative sentence on clause (ii) is correct because $W(N)/(N/\log N)\to\infty$. The Remark's attribution of what the weaker bound $\limsup\le1$ needs is imprecise (F1), and "Clause (i) remains open" is a status sentence on a page that changes no status (F2). Every consumed interface is supplied at the strength used; the unproved parts of the inputs (Theorem 3.1's sketch, Lemma 4.1's counting) are inherited and are disclosed on the page.
• Computation. Inapplicable. The page runs nothing; the decimal constants $C$ and $C'$ are quoted from p. 8 and are used only through $C>0$.
• Reproduction. Inapplicable. The page states no rerun command and cites no evidence folder.
• Source and verdict fidelity. Pass. The statement, the definitions, all locators (pp. 2, 5, 7, 8, 10 and §5), the quoted word "clearly" (p. 10), the p. 2 sentence that §3 alone gives $\limsup\le1$, the footnote that [8] means the corrected arXiv version (p. 7), and the bibliographic data of [4] and [8] (pp. 16--17) match the artifact; the page does not strengthen the source's conclusion, and its Standing sentence claims only an author-recorded reconstruction.

Weakest steps

W1: the uniform collision bound (B) for even $k$. Re-derived. Fix $\varepsilon(x)=(\log x)^{-1/2}$. Then $\varepsilon(x)\to0$ and $x^{\varepsilon(x)}=\exp(\varepsilon(x)\log x)=\exp(\sqrt{\log x})\to\infty$, so Theorem 3.3 (p. 6) applies to this function: there is a threshold $X_1$ such that for $x\ge X_1$ and every even $k$ with $2\le k\le x^{\varepsilon(x)}$, $P_0(x;k)\le(16C_2+1)c(k)x/(\log x)^2$, the "$+1$" absorbing the uniform $o(1)$. The range $1\le k\le\log x$ with $k$ even lies inside $[2,x^{\varepsilon(x)}]$ because $\log x\le\exp(\sqrt{\log x})$ is $\log t\le\sqrt t$ for $t=\log x>0$, which holds for every $t>0$ (the difference $\log t-\sqrt t$ is maximal at $t=4$, where it is negative). With $c(k)\le c^*$ absolute this gives $P_0(x;k)\le(16C_2+1)c^*x/(\log x)^2$. For $P_1$, Theorem 3.1 (p. 6) needs $k\le\exp((\log x)^{1/3})$, which holds once $\log\log x\le(\log x)^{1/3}$, true for large $x$; its bound $x/\exp((\log x)^{1/3})$ is at most $x/(\log x)^2$ once $(\log x)^{1/3}\ge2\log\log x$. For odd $k$ no $j$ has $\gamma(j)=\gamma(j+k)$ (exactly one of $j$, $j+k$ is even, and $j=1$ has no prime factor while $1+k\ge2$ has one), so $P_0(x;k)=0$. Adding, $P(x;k)\le(1+(16C_2+1)c^*)x/(\log x)^2$ for all natural $k\le\log x$ and all large $x$: this is (B) with $B=1+(16C_2+1)c^*$, as the page says. The absolute bound $c(k)\le c^*$ was re-derived from the source's displayed upper bound on p. 6: each factor $(p-1)/(p-2)$ is at most $(1-1/p)^{-2}$ (equality direction checked at $p=3$, and $(1+2/p)(1-1/p)^2=1-3/p^2+2/p^3\le1$ for $p\ge5$), so the product is at most $(k/\varphi(k))^2\ll(\log\log3k)^2$, while $7^{2\omega(k)}=k^{o(1)}$ from $\omega(k)\ll\log k/\log\log3k$; hence $c(k)\le3\cdot7^3k^{-1+o(1)}(\log\log3k)^2\to0$, and each $c(k)$ is finite, so $\sup_kc(k)<\infty$. This step composes with the main argument only through the single inequality $\#I_2\le\sum_{k\le\log x}P(x;k)\le Bx/\log x$.

W2: $x/\log x=o(W(x))$ from Ford's order of magnitude. Re-derived. $Z(x)/(x/\log x)=\exp(E(x))$ with $E(x)=C(\log_3x-\log_4x)^2+C'\log_3x-(C'+\tfrac12-2C)\log_4x$. Since $\log_4x/\log_3x\to0$, $(\log_3x-\log_4x)^2=(\log_3x)^2(1+o(1))$, so $E(x)=C(\log_3x)^2(1+o(1))+O(\log_3x)\to\infty$ because $C>0$. With $W(x)\asymp Z(x)$ for large $x$ (source p. 7), $W(x)/(x/\log x)\to\infty$. This step composes with the main argument by absorbing $1+(1+B)x/\log x$ into $o(1)\cdot W(x)$; it is also what makes the Remark's weaker bound $\limsup\le1$ work (see F1).

W3: the distinct-value indices and the assembly. Re-derived. List $S$ as $n_1<\cdots<n_m$, $k_i=n_{i+1}-n_i\ge1$, and split $\{1,\ldots,m-1\}$ into $I_1$ ($k_i>\log x$), $I_2$ ($k_i\le\log x$, equal totients) and $I_3$ ($k_i\le\log x$, unequal totients). Gaps: $\#I_1\log x<\sum_{i\in I_1}k_i\le n_m-n_1<x$. Repeats: for $i\in I_2$, $n_i\le x$ solves $\varphi(n)=\varphi(n+k_i)$ with a natural $k_i\le\log x$, and $i\mapsto n_i$ is injective, so $\#I_2\le\sum_{k\le\log x}P(x;k)\le\log x\cdot Bx/(\log x)^2$. Distinct: for $i\in I_3$ monotonicity gives $\varphi(n_i)<\varphi(n_{i+1})$, and for $i<i'$ in $I_3$, $n_{i+1}\le n_{i'}$ gives $\varphi(n_i)<\varphi(n_{i+1})\le\varphi(n_{i'})$, so $i\mapsto\varphi(n_i)$ is injective into $\varphi(S)\subseteq\mathcal W(x)$; by (A), $\#\varphi(S)=W(x)-\#(\mathcal W(x)\setminus\varphi(S))\le(1-c)W(x)$. Assembly: $m=1+\#I_1+\#I_2+\#I_3\le1+(1+B)x/\log x+(1-c)W(x)$, and for $m\le1$ the same bound holds trivially. Nothing in the three bounds depends on $S$ beyond its being nondecreasing in $[1,x]$, so the maximum over $S$ obeys the same bound, and W2 turns it into $(1-c+o(1))W(x)$, whence $\limsup\le1-c<1$.

Strongest attack

The strongest attempted refutation aimed at the uniformity behind (B), the one place where the page turns two asymptotic theorems with their own ranges into a single bound with absolute constants over a third range. Three angles were tried. (i) Theorem 3.3 is stated for one function $\varepsilon(x)$ at a time, and its $o(1)$ may depend on that function; the page fixes $\varepsilon(x)=(\log x)^{-1/2}$ before invoking it, so the threshold is a single absolute number, and the range $[2,\log x]$ sits inside $[2,x^{\varepsilon(x)}]$ for every $x>1$, so no $k$ escapes. (ii) The factor $c(k)$ could carry a hidden dependence on $k$ that grows along $k\le\log x$; the source's own upper bound (p. 6) and its Remark 3.1 give an absolute bound, and the re-derivation in W1 confirms that $c(k)\to0$. (iii) Odd $k$ lies outside Theorem 3.3 entirely; the page supplies the observation that $P_0(x;k)=0$ there, and the witness $j=1$ (no prime factors) was checked separately since "same prime factors" is vacuous for it. All three angles failed: the deduction of (B) is correct, and its only soft spot, the sketch status of Theorem 3.1, is inherited from the source and disclosed on the page and on the Theorem 3.1 page.

The attack that landed is not on the theorem's argument but on a consequence sentence. The Remark says the §3 collision bounds "alone" yield $\limsup M^\uparrow(x)/W(x)\le1$ and that Lemma 5.1 "or Ford's machinery ... enter only for the strict inequality". The "same argument" the Remark invokes uses W2, that is, input (C), to absorb $(1+B)x/\log x$: from (B) and the trivial $W(x)\ge\pi(x)$ alone one gets only $\limsup M^\uparrow(x)/W(x)\le1+(1+B)\limsup(x/\log x)/W(x)$, which is not $\le1$ without a lower bound on $W(x)$ that beats $x/\log x$. Under the reading "Ford's machinery = (C)" the last clause is wrong; under the reading "Ford's machinery = Ford's counting construction behind Lemma 4.1" it is right. The sentence is filed as F1 (suggested), a precision defect in a remark, not a defect in the reconstruction of the theorem.

Premises

• (A) Lemma 5.1. Interface as used: absolute $c>0$ and $x_0$ with $\#(\mathcal W(x)\setminus\varphi(S))\ge cW(x)$ for all $x\ge x_0$ and all nondecreasing $S\subseteq[1,x]$. Source held, p. 9, read on the page image; the source states it as "missing $\gg W(x)$ elements ... uniformly in the choice of $S$", which with the source's absolute implied constants (p. 3) is the interface. The input page's Statement is identical and its proof was read; that page is author-recorded and rests on Lemma 4.1, whose counting steps are Ford's and are not reconstructed.
• (B) collision bound. Deduced on the page from Theorem 3.1 (source p. 6, statement only; the source's proof is a sketch), Theorem 3.3 (source pp. 6--7, statement and proof read on the page images; the input page reconstructs it with Selberg's sieve, Evertse's bound and Theorem A imported) and Remark 3.1 (p. 6, $c(k)$ absolutely bounded; derived on the input page from two classical bounds cited to Hardy and Wright, not held). Interface: absolute $B$, $x_1$ with $P(x;k)\le Bx/(\log x)^2$ for $x\ge x_1$ and natural $k\le\log x$.
• (C) Ford's order of magnitude. Interface: $W(x)\asymp Z(x)$ for large $x$; quoted on source p. 7 from its [8], whose bibliography entry (p. 16, "Ramanujan J. 2 (1998), 67--151", cited as the arXiv version) matches the page. Ford's paper is outside this review's read set and was not read; the page says the same.
• (D) Erdős's count. Interface: $W(x)=x/(\log x)^{1+o(1)}$; quoted on source p. 2 from its [4], whose entry (p. 16, "Quart J. Math 6 (1935), 205--213") matches the page. Not read; used only for the consequence $M^\uparrow(x)=o(x)$, as the page says.
• Aside not checked. The sentence that Erdős's 1935 lower bound $W(x)\gg x\log_3x/\log x$ "as digested on the Erdős (1935) card, would serve equally" refers to a card outside the read set; the bound is implied by (C), but whether the card digests it from that paper was not verified. Nothing on the page depends on it.
• Problem 49 links. The clauses (i)--(iii) match the problem page's statement paragraph. The links to Tao's Theorem 1.1 and strict transfer pages resolve in the frozen state; their content is outside the read set and the page does not use it. The Lean name erdos_49 and the sentence about its proof script were not checked (the Formalization evidence section is outside the read set); the statement paragraph's own Formalization sentence agrees that the public target is the strict $o(N)$ clause.
• Explicit assumptions. $x$ real and large; $k$ a natural number; $n$ a positive integer; $S$ a set of integers in $[1,x]$; the thresholds "for large $x$" of Theorems 3.1 and 3.3 are absolute, as their statements say. No batch acceptance order applies.

Findings

F1. Severity: suggested. Location: Remark, "So the §3 collision bounds alone yield ... those enter only for the strict inequality." Defect: the sentence misattributes what the weaker bound $\limsup\le1$ consumes. The "same argument" it invokes uses $x/\log x=o(W(x))$, which the page derives from (C); without (C) the collision bounds give only $M^\uparrow(x)\le1+(1+B)x/\log x+W(x)$, and $W(x)\ge\pi(x)$ alone does not make the middle term $o(W(x))$. Witness: the page's own Conclusion step "using $x/\log x=o(W(x))$", and the (C) bullet, which says (C) is used exactly there; source p. 2 says only that §3 "already suffice[s]", with Ford's result treated as background on p. 10 ("clearly"). Proposed replacement: "So the §3 collision bounds, together with (C) through $x/\log x=o(W(x))$, yield $\limsup M^\uparrow(x)/W(x)\le1$, as the source notes on p. 2; the consequence $M^\uparrow(x)=o(x)$ below needs only (B) and (D). Lemma 5.1, and through it Lemma 4.1's use of Ford's construction, enter only for the strict inequality."

F2. Severity: suggested. Location: Consequence section, "Clause (i) remains open." Defect: a status sentence on a page whose Standing says it changes no status and whose warrant is the source's proof, which does not address clause (i). Witness: the source says nothing about strict extremality beyond p. 3, and the page cites no source for openness. Proposed replacement: "Clause (i) is not addressed by the theorem or by this page."

F3. Severity: note. Location: Main argument, "with $m=\#S\ge2$". Defect: the case $\#S\le1$ is dropped without a label; the final bound $\#S\le1+(1+B)x/\log x+(1-c)W(x)$ holds for it trivially, and the maximum over $S$ ranges over such sets. Witness: the sentence "Taking the maximum over $S$" covers the empty set and singletons. Proposed replacement: after "$m=\#S\ge2$" add "(for $\#S\le1$ the bound below is trivial)".

F4. Severity: note. Location: frontmatter title, "a nondecreasing totient set misses a fixed fraction of the totient values". Defect: the title paraphrases Lemma 5.1's conclusion (the image $\varphi(S)$ misses $\ge cW(x)$ values), not Theorem 1.2's (the size of $S$ is at most $(1-c+o(1))W(x)$); the two differ by the repeated-value and large-gap counts that the proof is about. Witness: the Lemma 5.1 page's title reads the same way. Proposed replacement: "Theorem 1.2: the nondecreasing totient maximum is a fixed fraction below the totient count".

F5. Severity: note. Location: Standing, "two of its inputs are only partly reconstructed", and (B), "deduced below from Theorem 3.1 and Theorem 3.3". Defect: the inventory of what is imported is incomplete on the page itself: (C) and (D) are unread external theorems, the deduction of (B) also uses Remark 3.1 (the absolute bound on $c(k)$, which the Theorem 3.3 page derives from two classical bounds not held), and $C_2$ is used without a local definition. Witness: the page's Imported inputs and Step 3. Proposed replacement: in Standing, "...only partly reconstructed (Theorem 3.1 is a sketch in the source; Lemma 4.1 imports Ford's counting argument), (C) and (D) are external theorems quoted from the source and not reread, and (B) also uses Remark 3.1 through the Theorem 3.3 page"; in Step 3, "where $C_2$ is the twin prime constant as normalized on source p. 5".

Verdict

Source fidelity: faithful. The statement, definitions, conventions, locators and both external citations match the artifact at the stated pages and labels, and the page claims only an author-recorded reconstruction.

The argument as reconstructed: sound, given its imported inputs at the depth their pages state. Every deduction on the page was re-derived above: (B) from Theorems 3.1 and 3.3 with Remark 3.1, $x/\log x=o(W(x))$ from (C), the three index classes and their bounds, the assembly, the uniformity in $S$, and the consequences for Problem 49. No required correction was found; two suggested corrections (F1, F2) concern a remark and a status sentence, and three notes (F3--F5) concern a trivial boundary case, the title and the inventory of imports.

Limitations: Theorem 3.1 is a sketch in the source, so (B) rests on a proof pointer; (C) and (D) are external theorems not read here; the aside on Erdős's 1935 lower bound and the Lean name erdos_49 were not verified; Tao's pages were not read. This is a focused fidelity and argument review of one page, not a review of its input pages.

This focused review assigns no tier and changes no status.