Skip to content

Documentary state: EDITORIALLY_ADMITTED. The manuscript has a protected editorial admission, not a theorem certification.

OAI-26 family 005: irrationality of Catalan's constant

Editorial status: DRAFT_UNACCEPTED. Mathematical status: NO_CERTIFICATION. This is a source-indexed explanation, not a verification of the proposed proof.

The question in plain language

Some familiar infinite sums have values whose arithmetic nature is surprisingly hard to determine. A number is rational if it equals a ratio of two integers with nonzero denominator; otherwise it is irrational.

Catalan's constant is a positive number built by alternately adding and subtracting reciprocals of odd squares. The OAI-26 Family 005 manuscript, dated September 24, 2026, claims to prove that this constant is irrational. This is an unusually strong assertion about a long-standing problem, and the existence of a manuscript claiming it is not evidence that the argument has been independently accepted.

A small example

The defining series is

\[ G=1-\frac{1}{3^2}+\frac{1}{5^2}-\frac{1}{7^2}+\cdots =\sum_{n=0}^{\infty}\frac{(-1)^n}{(2n+1)^2}. \]

The alternating-series test gives a simple numerical bound: after truncating at index \(N\), the absolute error is at most the magnitude of the next omitted term. So the first few terms yield increasingly accurate rational approximations.

But arbitrarily accurate rational approximations do not establish irrationality. Every real number, rational or irrational, admits such approximations. An irrationality proof needs an additional structural argument showing that no fixed fraction can equal the entire infinite sum.

For contrast, the irrationality of the square root of two can be proved by a divisibility contradiction. The same elementary contradiction does not automatically apply to an alternating infinite sum.

What is known

The publisher's pinned OAI-26 overview, Family 005, explicitly claims that the September 24 paper Catalan's constant is irrational proves irrationality of the constant defined above.

The pinned 44-page PDF states Theorem 1.1 on its first page. Its proposed argument is not simply a sequence of good rational approximants. It constructs mixed-moment determinants of size \(48N\) whose entries initially involve \(1\), Catalan's constant \(G\), and \(\zeta(2)\); a Taylor-contact condition is claimed to cancel the \(\zeta(2)\) contribution. If \(G\) were rational, the determinant entries would therefore be rational. The paper then compares finite-prime denominator bounds with a real-place upper bound, after separately seeking a nonzero determinant along an unbounded sequence of prime values of \(N\). The family has one indexed manuscript rather than companion papers.

The arithmetic problem should not be conflated with being able to compute many digits of the constant, with convergence of the defining series, or with proving that the number is transcendental. Irrationality means only that the value is not a fraction of integers; transcendence would assert more.

GCL verification state

The GCL documentary catalogue identifies exactly one manuscript under Family 005 and binds the source snapshot to the OpenAI mathematics repository revision 1581d4cbdd69a0faf34b2089614fe2a814316a6a. The family-level claim can be checked against the pinned public overview.

The exact September 24 pinned PDF was retrieved (Git blob ecbe60eb89d2d25c1d07da6232f8d9869e6c0c1d, 574,388 bytes, 44 pages), and its theorem statement and introductory proof architecture were inspected. The introduction (PDF p. 3) gives its proposed contradiction in terms of a normalized logarithm \(L_N\) of the determinant: Proposition 3.4 supplies a lower limit greater than \(-2.29084\) for nonzero determinants under rationality, Proposition 4.1 claims nonvanishing at unbounded prime indices, and Proposition 7.1 gives an upper limit at most \(-2.290939875\). If established with the same normalizations and quantifiers, those bounds would be incompatible. The evidence does not independently verify the determinant nonvanishing, denominator estimates, real-place energy bounds or the finite rational certificates. It has not identified a source-locked Lean theorem for Family 005, an associated Comparator replay, a validated paper-to-Lean semantic bridge, an adversarial mathematical audit, or a MATHCERT certificate. Corpus-wide formal-check statistics are not evidence for this particular statement without a documented binding.

The present artifact is an explanatory draft and must remain marked DRAFT_UNACCEPTED until a separately documented editorial decision is admitted.

What remains

A mathematical critic must verify the denominator lower bound (Proposition 3.4), prime-index determinant nonvanishing (Proposition 4.1), and energy upper bound with its rational certificates (Proposition 7.1), including whether all three concern the same normalized \(L_N\) and compatible subsequence. The paper describes a finite certificate for nonvanishing and rational trial-coefficient bounds; a reproducible replay of those arithmetic instructions is still required.

For a proof based on rational approximants, the critic should check denominators, error signs and bounds, and whether the contradiction works uniformly rather than only for numerical samples. These are diagnostic examples of obligations, not claims about which technique the author actually uses.

The documentary review should also compare the pinned source with later corrections, if any, without silently substituting a revised theorem for the locked one. A mathematical verdict must remain separate from editorial acceptance. After source-critical review, require valid Markdown and rendered HTML, a content-identified critic disposition, protected admission and readback before marking the exposition accepted.

Sources and claim boundaries

Boundary: NO_CERTIFICATION. The manuscript's claim of irrationality is being documented, not endorsed as a proved mathematical result by GCL. The article does not establish a novel proof, a verified semantic correspondence, or public certification.


Indexed primary manuscript paths: