gpt pro 5.4 attempt on Math-732

Erdős Problem #252

1) ROUND-2 OBJECTIVE

Path (C): keep the successful Round-1 factorial-tail framework, but sharpen it into an exact finite-dimensional reduction for the unresolved cases.

Round-1 already gave complete proofs for $k=1$ and $k=2$. The genuine remaining issue is the general case, where the naive prime-tail estimate is too crude. The most promising Round-2 move is therefore to determine exactly how many local terms one must keep, and exactly how small the remaining tail becomes.

2) Round-1 FOUNDATION USED

I use the following Round-1 results without re-proving them.

• If \[ \alpha_k:=\sum_{n=1}^{\infty}\frac{\sigma_k(n)}{n!} \] is rational and $N$ is at least the denominator, then the factorial tail \[ R_{N,k}:=\sum_{n=N+1}^{\infty}\sigma_k(n)\frac{N!}{n!} \] is an integer.
• The cases $k=1$ and $k=2$ were completely settled in Round-1.

In this round I work only with $k\ge 2$ and with the prime choice $N=p-1$.

3) NEW INSIGHT / TOOL (ROUND-2)

The new tool is an exact truncation threshold for the prime-tail method.

For a prime $p$ define \[ R_{p-1,k}=\sum_{j=0}^{\infty}U_{j,k}(p), \qquad U_{j,k}(p):=\frac{\sigma_k(p+j)}{p(p+1)\cdots(p+j)}. \] Then:

• If one truncates after fewer than $k$ local terms, the omitted tail is still at least $1$ (in fact usually much larger).
• If one keeps exactly the first $k$ local terms, the remaining tail is already $O_k(1/p)$.

Thus the irrationality problem for fixed $k\ge 2$ is reduced to understanding the near-integrality of the single finite expression \[ Q_{p,k}:=\sum_{j=0}^{k-1}U_{j,k}(p) \] along large primes $p$.

4) ATTACK PLAN (ROUND-2)

Round-1 identified the obstruction qualitatively: for $k\ge 3$ the early terms of $R_{p-1,k}$ are too large for the $k=1,2$ argument to work directly.

The exact claims to prove now are:

• a lower-bound statement showing that truncating after $L+1$ terms with $L\le k-2$ can never produce a subunit error term;
• an upper-bound statement showing that truncating after the first $k$ terms always leaves an error of size $O_k(1/p)$.

This turns the vague Round-1 obstruction into a sharp theorem: the first unresolved term count is exactly $k$.

5) WORK (ROUND-2)

Fix $k\ge 2$ and let $p$ be a prime.

Claim 1 (lower bound on each local term).

For every $j$ with $0\le j\le k-1$, \[ U_{j,k}(p)\ge p^{k-j-1}. \]

Proof.

Since $p+j$ is a divisor of itself, \[ \sigma_k(p+j)=\sum_{d\mid (p+j)} d^k \ge (p+j)^k. \] Also, \[ p(p+1)\cdots(p+j)\le p\,(p+j)^j. \] Therefore \[ U_{j,k}(p) =\frac{\sigma_k(p+j)}{p(p+1)\cdots(p+j)} \ge \frac{(p+j)^k}{p\,(p+j)^j} =\frac{(p+j)^{k-j}}{p} \ge p^{k-j-1}. \] $\square$

Corollary 2 (fewer than $k$ terms can never give a small remainder).

Let $L$ be an integer with $0\le L\le k-2$. Then \[ \sum_{j=L+1}^{\infty}U_{j,k}(p)\ge p^{k-L-2}. \] In particular, for every sufficiently large prime $p$ the remainder after truncating after the first $L+1$ terms is larger than $1$.

Proof.

The omitted tail is positive, so it is at least its first term: \[ \sum_{j=L+1}^{\infty}U_{j,k}(p)\ge U_{L+1,k}(p). \] Apply Claim 1 with $j=L+1$. $\square$

This proves a genuine barrier: the Round-1 style argument cannot work after keeping fewer than $k$ local terms.

Claim 3 (uniform $O_k(1/p)$ tail after $k$ terms).

Define \[ Q_{p,k}:=\sum_{j=0}^{k-1}U_{j,k}(p), \qquad E_{p,k}:=\sum_{j=k}^{\infty}U_{j,k}(p). \] Then for every prime $p\ge 2^k$, \[ 0<E_{p,k}\le \frac{2^{k+1}\zeta(k)}{p}. \]

Proof.

For $k\ge 2$ one has the standard bound \[ \sigma_k(n)=n^k\sum_{d\mid n}\frac1{(n/d)^k} \le n^k\sum_{m=1}^{\infty}\frac1{m^k} =\zeta(k)n^k. \] Hence, with \[ \nu_{j,k}(p):=\zeta(k)\frac{(p+j)^k}{p(p+1)\cdots(p+j)}, \] we have $U_{j,k}(p)\le \nu_{j,k}(p)$ for every $j$.

First, \[ \nu_{k,k}(p) \le \zeta(k)\frac{(p+k)^k}{p^{k+1}} \le \zeta(k)\frac{(2p)^k}{p^{k+1}} =\frac{2^k\zeta(k)}{p}, \] because $p\ge k$.

Second, for every $j\ge k$, \[ \frac{\nu_{j+1,k}(p)}{\nu_{j,k}(p)} =\frac{(p+j+1)^k}{(p+j)^k}\cdot\frac1{p+j+1} =\frac{(p+j+1)^{k-1}}{(p+j)^k} \le \frac{2^{k-1}}{p+j} \le \frac12, \] since $p\ge 2^k$.

Therefore the majorant tail is geometric: \[ E_{p,k} \le \sum_{j=k}^{\infty}\nu_{j,k}(p) \le 2\nu_{k,k}(p) \le \frac{2^{k+1}\zeta(k)}{p}. \] The positivity $E_{p,k}>0$ is obvious. $\square$

Corollary 4 (finite-dimensional reduction).

Assume $\alpha_k\in\mathbf{Q}$. Then for every sufficiently large prime $p$, \[ R_{p-1,k}\in\mathbf{Z} \] and hence \[ \min_{m\in\mathbf{Z}} \left| \sum_{j=0}^{k-1}\frac{\sigma_k(p+j)}{p(p+1)\cdots(p+j)}-m \right| \le \frac{2^{k+1}\zeta(k)}{p}. \]

Proof.

The Round-1 factorial-tail lemma gives $R_{p-1,k}\in\mathbf{Z}$ for every sufficiently large prime $p$. Since \[ R_{p-1,k}=Q_{p,k}+E_{p,k}, \] Claim 3 shows that the distance from $Q_{p,k}$ to the nearest integer is at most $E_{p,k}\le 2^{k+1}\zeta(k)/p$. $\square$

Interpretation.

The open part of Problem #252 is now reduced to the following explicit task:

For fixed $k\ge 3$, show that for infinitely many large primes $p$ the finite sum \[ Q_{p,k}=\sum_{j=0}^{k-1}\frac{\sigma_k(p+j)}{p(p+1)\cdots(p+j)} \] is not within $O_k(1/p)$ of any integer.

This is a strict strengthening of the Round-1 obstruction, because it identifies the exact number of local terms that must be controlled.

6) ADVERSARIAL VERIFICATION

• Claim 1 only uses the trivial divisor $d=p+j$ and the elementary denominator bound \[ p(p+1)\cdots(p+j)\le p(p+j)^j. \] No hidden hypothesis is present.
• Corollary 2 is sharp in the sense relevant here: when $L=k-2$, the omitted tail is still at least $1$, so no interval of length $<1$ can arise from such a truncation.
• Claim 3 uses only the convergent Euler sum $\zeta(k)$ and a geometric-ratio estimate valid for every $p\ge 2^k$.
• The reduction in Corollary 4 is fully rigorous and depends only on the Round-1 integer-tail lemma.

Thus there is no logical gap in the new reduction: the unresolved issue is purely the arithmetic behavior of the explicit $k$-term local sum $Q_{p,k}$.

7) FINAL

UNRESOLVED (BUT STRICTLY ADVANCED).

The strict Round-2 advance is an exact threshold theorem for the prime-tail method.

• Truncating after fewer than $k$ local terms can never leave a subunit remainder.
• Truncating after exactly the first $k$ local terms always leaves an error $O_k(1/p)$.
• Therefore rationality of $\alpha_k$ would force the explicit $k$-term quantity \[ Q_{p,k}=\sum_{j=0}^{k-1}\frac{\sigma_k(p+j)}{p(p+1)\cdots(p+j)} \] to be $O_k(1/p)$-close to an integer for every sufficiently large prime $p$.

This is a genuine finite-dimensional reduction of the unresolved cases.

8) COMPLETION ESTIMATE (MANDATORY)

COMPLETION: 55%

9) REFERENCES

Only references actually used or checked.

• T. F. Bloom, Erdős Problem #252, https://www.erdosproblems.com/252, accessed 2026-03-07.
• P. Erdős, Problem 4493, Amer. Math. Monthly 59 $1952$, 557--558.
• J.-C. Schlage-Puchta, The irrationality of a number theoretical series, Ramanujan J. 15 $2008$, 149--158; arXiv version checked at https://arxiv.org/abs/1105.1452.
• J. B. Friedlander, F. Luca, and M. Stoiciu, On the irrationality of a divisor function series, Integers 7 $2007$, A36.
• K. Pratt, The irrationality of a divisor function series of Erdős and Kac, arXiv:2209.11124.

---