Skip to content

Erdős Problem #251

Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: open (not automatically the live result).
Subject areas: number theory, irrationality.

Read this status correctly

The status above is reproduced from GCL's immutable external-catalogue snapshot. It is not a new claim of current mathematical resolution, a source-semantic review, an independent proof check, or GCL certification. A published proof and a GCL replay are separate events.

Documentary explanation. Editorial acceptance is determined by the protected V2 admissions ledger and the exact manuscript source hash. This article does not confer mathematical certification.

The question in plain language

Let \(p_n\) be the prime indexed by \(n\), so \(p_1=2\), \(p_2=3\), and so on. Erdős Problem 251 asks about the single, well-defined real number

\[ S=\sum_{n=1}^{\infty}\frac{p_n}{2^n}. \]

Is \(S\) irrational? The denominator doubles at every step, so the series converges even though the primes grow. The question is not whether every series built from primes has this property, nor whether a numerical decimal expansion looks nonrepeating. It asks for a proof that this particular real number cannot be written as a quotient of two integers.

There is a useful nearby contrast. Erdős proved that

\[ \sum_{n=1}^{\infty}\frac{p_n^k}{n!} \]

is irrational for every integer \(k\ge1\). That classical theorem has a faster factorial denominator and does not settle Problem 251, whose denominator is \(2^n\).

A small example

The first ten primes give a completely checkable finite approximation. With

\[ (p_1,\ldots,p_{11})=(2,3,5,7,11,13,17,19,23,29,31), \]

exact rational arithmetic gives

\[ S_{10}=\sum_{n=1}^{10}\frac{p_n}{2^n}=\frac{3727}{1024}. \]

The same data illustrate a finite summation-by-parts identity. If \(\delta_n=p_{n+1}-p_n\) and

\[ D_{10}=\sum_{n=1}^{10}\frac{\delta_n}{2^n}, \]

then

\[ D_{10}=\frac{855}{512}, \qquad S_{10}=2+D_{10}-\frac{31}{2^{10}}. \]

An independent reproduction uses only Python's standard-library Fraction type:

from fractions import Fraction

primes = [2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31]
S10 = sum((Fraction(p, 2**n) for n, p in enumerate(primes[:10], start=1)), Fraction())
D10 = sum((Fraction(primes[n] - primes[n-1], 2**n) for n in range(1, 11)), Fraction())
assert S10 == Fraction(3727, 1024)
assert D10 == Fraction(855, 512)
assert S10 == 2 + D10 - Fraction(primes[10], 2**10)
print(S10, D10)

The output is 3727/1024 855/512. This verifies a finite identity and the supplied prime data only. It neither computes the infinite tail nor gives evidence that \(S\) is rational or irrational.

What is known

The primary sources separate a proved neighbouring result from the unresolved target. In the 1958 paper, Erdős proves irrationality for the factorial-denominator series above and then explicitly writes that he had not succeeded in proving irrationality for the series with denominator \(2^n\) and numerator \(p_n\). The 1988 survey repeats the distinction: it records the factorial result, says that the \(p_n^k/2^n\) series was not proved irrational for every \(k\), and describes the \(k=1\) case as probably very difficult. It then states the broader conjecture that if \(g_n\ge2\) and \(g_n=o(p_n)\), then

\[ \sum_{n=1}^{\infty}\frac{p_n}{g_1g_2\cdots g_n} \]

should be irrational. This broader conjecture contains the spirit of the problem but is not a proof of it.

The maintained Erdős Problems entry currently presents Problem 251 as open, cites the 1958 paper, Erdős and Graham's 1980 book, and the 1988 survey, and reports that the page was last edited on 28 September 2025. Its statement metadata says that a formalized statement exists; that is not a solved-status claim. No externally reported proof of the specific \(\sum p_n/2^n\) assertion was identified in the audited sources.

GCL's current state

The authenticated RESULT/1 audit on MATH-PROGRAMME issue #1521 has disposition PARTIAL. It records the pinned source statement, the live-page check, and an exact reduction, but it does not certify a solution.

For that reduction, set \(\delta_n=p_{n+1}-p_n\) and

\[ G_N=\sum_{j=0}^{\infty}\frac{\delta_{N+j}}{2^j}. \]

Separating the first \(N\) terms and regrouping the tail gives

\[ \{2^N S\}=\{G_N\}, \]

where braces denote fractional part. A rational real number has an eventually periodic sequence of fractional parts under repeated doubling. Consequently, if \(S\) were rational, then \((\{G_N\})_{N\ge1}\) would be eventually periodic. In particular, for every proposed period \(m\), all sufficiently late \(N\) would satisfy

\[ (2^m-1)G_N\in\mathbb Z. \]

Showing infinitely many failures of that integer condition for every \(m\) would prove irrationality. This is an exact necessary-condition route, not an established nonperiodicity theorem.

The pinned formal file uses a zero-based natural-number index and contains answer(sorry). The audit correctly retains semantic-review and reindexing caveats: the file's existence demonstrates formal statement capture, not a proof, kernel certification, or semantic-equivalence certification by GCL.

What remains

The central unresolved task is to rule out eventual periodicity of the dyadic tails of the prime-gap sequence, or to find a different proof that \(S\) is irrational. A finite computation can certify selected pairs \((m,N)\) by evaluating an initial segment exactly and enclosing the remaining tail in a rational interval that avoids integers. It cannot, by itself, handle every \(m\) and every sufficiently large \(N\).

The audit's proposed bounded experiment therefore has a clear stopping rule: for fixed finite bounds, certify all selected interval exclusions or report the first failure, while retaining the exact prime fixture, tail theorem, interval endpoints, and input hashes. Either outcome is local evidence. Neither outcome resolves the unbounded problem without a uniform argument. Conversely, the factorial-denominator theorem, the decimal expansion, and the existence of a formalized statement must not be promoted into a result for \(S\).

Sources and claim boundaries

The scan hashes identify the downloaded bytes and preserve the audit trail; they do not independently certify publisher provenance. During local retrieval the scan host required a TLS trust bypass, so that access limitation is retained rather than hidden. No GCL theorem, independent reviewer signoff, external proof, formal certification, or publication authority is asserted here.

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1521 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
  • GCL triage: PROTECTED_READY_BACKLOG.
  • Formal-statement coverage: A provider statement was included in the pinned snapshot. Presence does not mean the statement or proof has been checked.
  • External mathematical status: consult current problem page and its dated bibliography.

Back to catalogue · Compendium introduction