---

Subject and independence

Role: an independent reviewer in a fresh context, commissioned for refutation, who took no part in writing the page and read only the material listed here. The commission fixed the read set; no other review, assessment or status text was consulted beyond the exposures disclosed below.

Frozen subject: wiki/research/erdos_15/relation_2_1_reconstruction.md as it stood on 2026-09-28T05:03:27Z, read from the committed text. The working tree sat at that state during the review, and the PDF named below is identical in that state and in the tree.

Artifact: the sixteen-page arXiv v3 PDF (23 August 2023) held by Tao (2023). Physical pages 1--4 were read in full from a layout text extraction: page 1 for Questions 1.1 and 1.2, footnote 3 and the "convenience of the reader" sentence; page 2 for the statement of Theorem 1.4; page 3 for the asymptotic notation convention and the start of Section 2; page 4 for the rest of Section 2, Remark 2.1 and footnote 4. Physical pages 3 and 4 were also rendered as page images (page 3 at 110 dpi, page 4 at 110 and 160 dpi), and every display of Section 2 was read from the images: (2.1), the shift display, the averaging display, (2.2), the alternating-series display, the subdivision display, the difference series, (2.3), the intermediate value display, the prime number theorem form, the comparison display and the summation-by-parts display. Physical and printed page numbers agree (the running heads print 3 and 4). Pages 5--16 were not read. Section 2 of the canonical conversion beside the PDF was read and agrees with the PDF; the PDF decided.

Allowed material read: the provenance paragraph of the card _index.md in the same folder (the card extracts no result pages, so none were read); the Statement of Problem 15; docs/verification.md, sections "Whole-claim report" and "Audit checklist"; docs/evidence.md, section "Source fidelity"; docs/math_authoring.md in full; and, to check the cross-link in the page's Boundary paragraph, the Statement section of the Theorem 1.4 reconstruction and the sentence of its Source paragraph that links this page. The second library card named in the commission (kuperberg_2025_alternating_series_primes) does not exist in that state and is not linked by the page; nothing was read from it or any other card.

Exposures: three, all incidental and none used in the verdict. (1) The first sixty lines of the Problem 15 page were printed to reach its Statement, which put its Status paragraph and its proof-file provenance paragraph in view. (2) The whole body of the Tao card was printed rather than its provenance paragraph alone, including its Bears-on line, its results-to-transcribe list and its closing paragraph naming the reconstruction pages. (3) The Standing paragraph of the Theorem 1.4 reconstruction was printed together with its Source paragraph. Nothing among the private working files, no evidence folder, no other review and no web search was consulted. A short numeric sanity check over the primes below $1.2\times10^7$ was run; it is described under Strongest attack and is not evidence.

Restatement

Let $p_n$ be the $n$th prime and $\pi(t)$ the number of primes at most $t$. For real $x\ge1$ and real $y\ge2$ write

$$A(x)=\sum_{n\le x}\frac{(-1)^nn}{p_n}, \qquad B(y)=\sum_{2\le m\le y}\frac{(-1)^{\pi(m)}}{m\log m}.$$

The claim, the source's (2.1): there is one real number $C$, depending on nothing, such that $A(x)-\tfrac12B(x\log x)-C\to0$ as the real variable $x\to\infty$. The consequence, the source's equivalence of its Questions 1.1 and 1.2: the limit of $A(x)$ as $x\to\infty$ exists if and only if the limit of $B(y)$ as $y\to\infty$ exists; in series form, the series of Problem 15 converges if and only if the series of the parity of the prime counting function over $n\log n$ converges. Conventions: $\log$ is the natural logarithm; $O(\cdot)$, $\ll$ and $\asymp$ carry absolute implied constants (source p. 3); $o(1)$ is a quantity tending to zero as $x\to\infty$. The one imported input is the prime number theorem in the form $p_n=n\log n\,(1+O(\log\log n/\log n))$ for $n\ge10$. No conjecture is assumed; the result is unconditional.

Checklist

The audit checklist lists seven canonical failure modes and twelve named patterns; each is given a verdict here.

Canonical failure modes.

• "Almost all" upgraded to "all": inapplicable. The page makes no density or almost-all statement; every asymptotic holds for all $n\ge10$ or all large real $x$.
• Induction presupposing termination: inapplicable. No induction is used.
• Probabilistic or averaging heuristics presented as proofs: pass. The one averaging step (Step 1) is an exact identity, the mean of the two equal expressions for $A(x)$; no heuristic enters.
• Circular use of a statement equivalent to the claim: pass. Neither series is assumed convergent anywhere; Step 5 proves the absolute convergence of the difference series from the prime number theorem alone, and Step 9 derives each direction of the equivalence from (2.1).
• Exceptional sets dropped from density arguments: inapplicable. Every $o(1)$ is a genuine limit along all real $x\to\infty$, not a bound off an exceptional set.
• Finite verification cited as more than base-case coverage: pass. The page cites no computation. My own numeric check (Strongest attack) is used only as an attack, not as support.
• Convergence of a relaxed or averaged system standing in for the actual objects: pass. The relation between the actual partial sums is exact up to $o(1)$; the absolutely convergent series of Step 5 is the actual difference of the actual summands.

Named patterns.

• Model-class transport instead of entailment: inapplicable; no axiom system or certificate class is classified.
• Uniformity over an infinite family asserted from finitely many instances: pass. Every $O$ and $\ll$ on the page is derived analytically for all $n\ge10$ (or all $t\ge10$) with absolute constants, never verified on instances; the constants $C$, $C_0$ and $D$ are single real numbers.
• Extremal claims audited in the claim's own units: inapplicable; no sharpness, supremum or attainment sentence is claimed.
• Consequence sentences are claim surfaces: pass. The "Consequently" sentence of the Statement was attacked on its own (Step 9, and the direct route in F4); both directions follow from (2.1).
• Carry hypotheses actually used by a quantified argument: pass. The only hypothesis used is the imported prime number theorem form, and it is stated at the strength consumed with its range $n\ge10$.
• A composition inherits its unproved premises: pass. The one premise is named as imported and not reproved; the page's Standing claims unconditionality only modulo that classical theorem, which is correct.
• Reproducibility notes are claims: inapplicable; the page carries no rerun line, check count or harness statement.
• Verifier quotations are claims: inapplicable; the page quotes no verifier and claims no review.
• Verdict words spelled in full: inapplicable to the page; this report writes its verdicts in full.
• Certified-bracket functions fail loudly: inapplicable; no numeric routine is part of the page.
• A harness leg with no failing input is decoration: inapplicable; no harness.
• A gate that reads caches instead of re-running is defective: inapplicable; no gate rides on the page.

Weakest steps

Step 4: replacing $B(p_{\lfloor x\rfloor+1}-1)$ by $B(x\log x)$

Re-derivation. For $n=1,\dots,\lfloor x\rfloor$ the intervals $[p_n,p_{n+1})$ are disjoint with union $[2,p_N)$, $N=\lfloor x\rfloor+1$, so the double sum over $n\le x$ and $p_n\le m<p_{n+1}$ is exactly $B(p_N-1)$. From the imported form, $p_N=N\log N\,(1+o(1))$, and $N\log N=x\log x\,(1+O(1/x))$ since $N\in(x,x+1]$; hence $p_N-1=x\log x\,(1+o(1))$. Let $a\le b$ be the two numbers $x\log x$ and $p_N-1$ in order; then $b/a\to1$. With $g(t)=1/(t\log t)$, positive and decreasing on $(1,\infty)$, the integers $m\in(a,b]$ contribute $|B(b)-B(a)|\le\sum_{a<m\le b}g(m)$; the least such $m$ gives at most $g(a)$, and each later one at most $\int_{m-1}^mg$, so the sum is at most $g(a)+\int_a^bg=g(a)+\log(\log b/\log a)$. Since $\log(\log b/\log a)=\log\bigl(1+\log(b/a)/\log a\bigr)\le\log(b/a)/\log a$ and $\log(b/a)\to0$, both terms are $o(1)$. Composition: this is exactly the source's unnumbered subdivision display on p. 4 with its $o(1)$ made explicit; it feeds Step 5, where the double sum is traded for $B(x\log x)$ at the cost of one more $o(1)$.

Step 7: the comparison of $1/(x_n\log x_n)$ with $n/(p_np_{n+1})$

Re-derivation. Write $\eta_n=\log\log n/\log n$, positive for $n\ge10$. The imported form gives $p_n=n\log n\,(1+O(\eta_n))$ and, at index $n+1$, $p_{n+1}=(n+1)\log(n+1)\,(1+O(\eta_{n+1}))$; since $(n+1)\log(n+1)-n\log n=\log(n+1)+n\log(1+1/n)=\log n+O(1)$, the ratio $(n+1)\log(n+1)/(n\log n)$ is $1+O(1/n)$, and $\eta_{n+1}\asymp\eta_n$, so $p_{n+1}=n\log n\,(1+O(\eta_n))$. The point $x_n$ lies in $[p_n,p_{n+1})$, so $x_n=n\log n\,(1+O(\eta_n))$, and taking logarithms, $\log x_n=\log n+\log\log n+O(\eta_n)=\log n\,(1+O(\eta_n))$, because $\log\log n=\eta_n\log n$. Hence $x_n\log x_n=n\log^2n\,(1+O(\eta_n))$ and $p_np_{n+1}=n^2\log^2n\,(1+O(\eta_n))$; both are positive, so their reciprocals are $(1+O(\eta_n))$ times $1/(n\log^2n)$ and $1/(n^2\log^2n)$, the inversion being legitimate for large $n$ because $\eta_n\to0$ and for the finitely many $10\le n\le n_0$ because the quantities are positive and finite in number. The difference of two quantities of the form $(1+O(\eta_n))/(n\log^2n)$ is $O(\eta_n/(n\log^2n))=O(\log\log n/(n\log^3n))$. This is the source's comparison display on p. 4. Composition: multiplied by the gap $p_{n+1}-p_n$ (Step 6), it bounds the $n$th term of (2.3) by a constant times $w_n(p_{n+1}-p_n)$, $w_n=\log\log n/(n\log^3n)$, which Step 8 sums.

Step 8: summing $w_n(p_{n+1}-p_n)$

Re-derivation. Abel summation:

$$\sum_{n=10}^Nw_np_{n+1}-\sum_{n=10}^Nw_np_n =\sum_{n=11}^{N+1}w_{n-1}p_n-\sum_{n=10}^Nw_np_n =w_Np_{N+1}-w_{10}p_{10}-\sum_{n=11}^N(w_n-w_{n-1})p_n,$$

as on the page. For $w(t)=\log\log t\cdot t^{-1}(\log t)^{-3}$ the product rule gives $w'(t)=\frac{1}{t\log t}\cdot\frac{1}{t\log^3t}-\log\log t\cdot\frac{\log t+3}{t^2\log^4t}$, the page's formula. For $t\ge10$, $\log t\log\log t\ge\log10\cdot\log\log10>1.9$, so the first term is at most the second term's size, and $|w'(t)|\ll\log\log t/(t^2\log^3t)$. By the mean value theorem on $[n-1,n]$, $n\ge11$, $|w_n-w_{n-1}|\ll\log\log n/(n^2\log^3n)$. With $p_n\ll n\log n$ from the imported form, $\sum_{n\ge11}|w_n-w_{n-1}|p_n\ll\sum_{n\ge11}\log\log n/(n\log^2n)$, and under $u=\log\log t$ (so $\log t=e^u$, $du=dt/(t\log t)$) the integrand $\log\log t/(t\log^2t)$ becomes $ue^{-u}$, whose integral converges; the summand is eventually decreasing, so the integral test applies. The boundary term $w_Np_{N+1}\ll\log\log N/\log^2N\to0$. So the partial sums converge, the terms are nonnegative, and the series converges. This matches the source's summation-by-parts display on p. 4, whose "$1+$" absorbs the two boundary terms. Composition: it proves (2.3), hence the absolute convergence hypothesized in Step 5, hence (2.1).

Strongest attack

The strongest attack aimed at what the page adds to the source: the explicit constant $C=-\tfrac14-\tfrac{C_0}2-\tfrac D2$ of Step 5, which the source never writes, and the claim that the difference $A(x)-\tfrac12B(x\log x)$ converges at all. If the bookkeeping of the $o(1)$ terms in Steps 1--5 lost a boundary term or a factor $\tfrac12$, the page would still read smoothly but (2.1) would hold with the wrong constant or not at all. I recomputed the chain independently: Step 1's reindexing has boundary terms $-(-1)^1\cdot1/p_1=+\tfrac12$ and $T=(-1)^NN/p_N$ with $N=\lfloor x\rfloor+1$, so $A(x)=-\tfrac12+\sum_{n\le x}(-1)^{n+1}(n+1)/p_{n+1}-T$; averaging with $A(x)$ gives $-\tfrac14+\tfrac12\sum(\cdots)-T/2$; (2.2) is an exact identity (checked by clearing denominators: $np_{n+1}-(n+1)p_n=n(p_{n+1}-p_n)-p_n$); the alternating-series constant $C_0$ and the absolutely convergent sum $D$ enter with the factor $\tfrac12$ from the averaging. The constant is therefore $-\tfrac14-\tfrac{C_0}2-\tfrac D2$. As a sanity check, not as evidence, a short script over the primes below $1.2\times10^7$ (so $n$ up to about $7.9\times10^5$) gave $C_0\approx-0.2304$, $D\approx-0.3725$ with the absolute series converging to about $1.078$, hence $C\approx0.0514$; the observed $A(x)-\tfrac12B(x\log x)$ at $x=6\times10^5$ was about $0.0849$, and the gap $0.0335$ is exactly $-T/2$ for $N=600001$ ($T\approx-0.067$), the slowest $o(1)$ in the chain, of size about $1/(2\log x)$. The same script confirmed (2.2) and the Abel identity to rounding, the intermediate value bracket $g(p_{n+1}-1)\le$ average $\le g(p_n)$ for $n\le2000$, and that the Step 7 error divided by $\log\log n/(n\log^3n)$ never exceeded $0.40$ for $10\le n<788058$. The attack failed: the constant and the $o(1)$ structure are as the page states.

Secondary attacks, all failed: (i) the mean-value point when the gap is $1$ ($n=1$, $p_1=2$, $p_2=3$): the interval $[p_n,p_{n+1}-1]$ is the point $2$ and the identity is trivial, so Step 6 needs no exception; (ii) the inversion $1/(1+O(\eta_n))=1+O(\eta_n)$ in Step 7 for small $n$, where the implied constant of the prime number theorem could make $1+O(\eta_n)$ small: positivity of $x_n\log x_n$ and $p_np_{n+1}$ and finiteness of the range $10\le n\le n_0$ make the constant absolute; (iii) the converse in Step 9, attacked by asking whether integer $x$ and a counting argument suffice for real $y$: they do, and the page's bound $|B(y)-B(x\log x)|\ll1/(x\log x)$ follows from at most $\log(x+1)+2\ll\log x$ integers in $(x\log x,y]$, each of size at most $1/(x\log x\log(x\log x))$; (iv) the tiling in Step 4 at the left end: $[p_1,p_2)=[2,3)$ starts $B$ at $m=2$ as required.

Premises

• Prime number theorem, in the form $p_n=n\log n\,(1+O(\log\log n/\log n))$ for all $n\ge10$ with an absolute implied constant; consequences used: $p_n\ll n\log n$, $p_{\lfloor x\rfloor+1}=x\log x\,(1+o(1))$, $n/p_n\to0$. Held source for this form: the source itself states it on p. 4 ("From the prime number theorem we have ...") without proof; read in full at that display. The page derives it from $\pi(t)=\frac t{\log t}(1+O(1/\log t))$, a classical statement with no held source; the derivation was checked (see F3) and is correct. Standing on the page: named as imported, which is right; it is not reproved.
• Elementary analysis used within its hypotheses: the alternating series test ($1/p_{n+1}$ decreasing to $0$); the intermediate value theorem for the continuous decreasing $g(t)=1/(t\log t)$ on $[p_n,p_{n+1}-1]$; the mean value theorem for $w$ on $[n-1,n]$, $n\ge11$; Abel summation; the integral test for the eventually decreasing $\log\log t/(t\log^2t)$. These are not number-theoretic imports and have no held source; none is applied outside its hypotheses.
• Local claims consumed: none. Sibling reconstruction pages consumed: none; the Theorem 1.4 reconstruction is a consumer of this page, as its Statement confirms ("hence, by relation (2.1)"), and the page's Boundary sentence describing that use is accurate.
• Explicit assumptions: none beyond the prime number theorem. Batch acceptance order: not applicable.

Findings

F1

Severity: suggested.

Location: "Standing." paragraph, and Steps 4, 5, 7, 8 and 9 together with "Imported input (prime number theorem)".

Defect: the page does not mark which parts of the proof are the source's and which are supplied by the reconstruction. The source (pp. 3--4) writes the displays (2.1)--(2.3) and the intermediate displays reproduced in Steps 1--4, 6 and 7, but gives the tail estimate of Step 4 only as "from the prime number theorem and subdivision of the $m$ variable", the calculation of Step 7 only as "after some calculation", the derivative bound and the integral comparison of Step 8 only as "from summation by parts and the prime number theorem", the equivalence of Step 9 only as "clearly follows", and never writes the explicit constant $C=-\tfrac14-\tfrac{C_0}2-\tfrac D2$ of Step 5 nor the derivation of the $p_n$ form from $\pi(t)$. All of these supplied pieces are correct, but a reader cannot tell from the page where the source stops and the reconstruction starts, which the commission's labeling rule requires.

Witness: source p. 3, the sentence before (2.2) and the alternating series sentence; source p. 4, the sentences "from the prime number theorem and subdivision of the $m$ variable", "and thus after some calculation", "from summation by parts and the prime number theorem we have" and "and the claim follows"; source p. 3, "from which the equivalence of the two questions clearly follows".

Proposed replacement: add to the Standing paragraph the sentence "The source states displays (2.1)--(2.3) and the intermediate displays reproduced in Steps 1--4, 6 and 7; the tail estimate in Step 4, the explicit constant in Step 5, the calculation in Step 7, the derivative bound and integral comparison in Step 8, the derivation of the prime number theorem form from $\pi(t)$, and Step 9 are supplied by this reconstruction where the source writes 'after some calculation', 'from summation by parts and the prime number theorem' and 'clearly follows'." Alternatively mark each supplied passage in place with "(supplied)".

F2

Severity: note.

Location: "Boundary." paragraph, "can be taken to be $O(\log\log x/\log x)$"; and Step 5, "The starting index $10$ is arbitrary (the source's footnote 4)".

Defect: two paraphrases drop a qualification. Remark 2.1 states its bound "for $x\ge10$"; the page omits the range. Footnote 4 says the index $10$ is chosen "rather arbitrarily; any index for which $\log\log n$ is well-defined and positive would suffice", so the admissible indices are $n\ge3$, not arbitrary; the page's "arbitrary" loses the condition. Neither affects the proof, since Remark 2.1 is not reconstructed and the page handles the first nine terms directly.

Witness: source p. 4, Remark 2.1 and footnote 4.

Proposed replacement: "that the $o(1)$ in (2.1) can be taken to be $O(\log\log x/\log x)$ for $x\ge10$" and "The starting index $10$ is the source's choice (its footnote 4: any index with $\log\log n$ defined and positive would do); the first nine terms are finite."

F3

Severity: note.

Location: "Imported input (prime number theorem)", the sentence "it gives $p_n=n\log p_n\,(1+O(1/\log n))$, and $\log p_n=\log n+O(\log\log n)$".

Defect: the second clause silently uses an a priori bound $\log p_n\ll\log n$; without it, $\log\log p_n$ is not visibly $O(\log\log n)$. The bound does follow from the first clause: taking logarithms, $\log p_n-\log\log p_n=\log n+O(1)$, and $u-\log u\ge u/2$ for $u\ge4$, so $\log p_n\le2\log n+O(1)$ and $\log\log p_n\le\log\log n+O(1)\ll\log\log n$ for $n\ge10$. The step is correct but one link is missing from a derivation the page supplies (see F1).

Witness: the page's own display; the source (p. 4) states the $p_n$ form without deriving it, so there is no source witness to compare.

Proposed replacement: "it gives $p_n=n\log p_n\,(1+O(1/\log n))$, hence $\log p_n-\log\log p_n=\log n+O(1)$, so $\log p_n\ll\log n$ and $\log p_n=\log n+O(\log\log n)$."

F4

Severity: note.

Location: Step 9, "suppose $A(x)$ converges as $x\to\infty$ through the integers. By (2.1), $B(x\log x)$ converges along the integers $x$."

Defect: none in validity; a reading is introduced that the argument does not need. Steps 1--8 prove (2.1) for real $x\to\infty$, and $x\mapsto x\log x$ is a continuous increasing bijection of $[1,\infty)$ onto $[0,\infty)$, so for real $y\to\infty$ the real $x$ with $x\log x=y$ tends to infinity and $B(y)=2A(x)-2C+o(1)$ converges directly; the integer restriction and the count of integers in $(x\log x,y]$ are correct but superfluous. Recording this keeps a future reader from suspecting a gap between integer and real limits.

Witness: the page's Statement, "$(x\to\infty)$" over real $x$, and Steps 1 and 4, which use $\lfloor x\rfloor$ for real $x$.

Proposed replacement: "Conversely, suppose $A(x)$ converges as $x\to\infty$. For real $y\ge0$ let $x\ge1$ be the real number with $x\log x=y$; then $x\to\infty$ with $y$, and (2.1) gives $B(y)=2A(x)-2C+o(1)$, which converges." The existing counting argument may stay as a remark.

Verdict

Source fidelity: faithful. The Statement reproduces (2.1) exactly, including the range $2\le m\le x\log x$, the absolute constant and the $o(1)$ convention; the Definitions match Questions 1.1 and 1.2; the credit to the unpublished observation, the footnote 3 pointer to MathOverflow question 313999 and the quoted phrase match p. 1; the locators (Section 2, displays (2.1)--(2.3), physical and printed pp. 3--4, sixteen-page arXiv v3) are correct; the displays reproduced in Steps 1--4, 6 and 7 agree with the source displays sign for sign. The findings above do not alter any statement the source proves.

The argument as reconstructed: sound. Every deduction in Steps 1--9 was re-derived and holds; the one imported theorem is stated at the strength consumed, applied within its range $n\ge10$, and named as imported.

Limitations: the review covers the page and pp. 1--4 of the source only; the prime number theorem was accepted as a classical import and not traced to a held proof; Remark 2.1 was not examined beyond confirming that the page declines to reconstruct it; the numeric check is a sanity attack over a finite range and warrants nothing. This focused review assigns no tier and changes no status.