Auditing a manuscript on the number of distinct totients
One audit, start to verdict, of a manuscript on how many distinct values Euler's phi function takes. Every finite check passed, but the general theorem is unverified and the printed pages are still unread.
This is one audit from start to verdict. The finite checks passed, the general theorem is unverified, and the PDF's printed pages have not been read. It was a run at 05:35 on 16 October, and the steps below are in the order it took them.
Euler's totient φ(n) counts the numbers up to n that share no factor with n. RSA arithmetic is built on it. The manuscript, An asymptotic formula for the number of totients (OpenAI, dated 25 September 2026), asks how many distinct values φ takes up to x.
1. The exact file is pinned but no printed page was read
The upstream repository can change, so the run recorded the exact commit (a fixed version of the repository) and the Git blob of the PDF (a fingerprint of the file's exact contents). The PDF came back as encoded text and was never rendered, so the audit states that no printed page was read. Page numbers in the notes are labelled a provisional secondary map.
2. The sharper formula is what is new
The original LaTeX has seven substantive sections; the longest are extraction (704 lines) and limits (516). The central claim, Theorem 2.1, is that the count V(x) behaves like x times an explicit factor over log x, with a coefficient that depends on a phase θ, and that V(cx)/V(x) tends to c. The run marks it claimed.
Earlier work (Ford, 1998 and 2013; Maier and Pomerance, 1988) gave the order of magnitude. The sharper equivalence is what is new, if it holds.
3. Five places it could break
The five are: the extraction step that imports Ford's results for every preimage; the sieve (a counting method that strikes out multiples) with coefficients up to y; the collision count, which counts pairs of tuples and relies on a tiny strict saving; the coefficient step, where the order of two limits cannot be swapped without new proof; and the companion result on least preimages.
The finite-map counting lemma was tested in 3,000 randomised cases. The analytic collision estimate was not.
4. 9,506 finite checks with no failures
A sieve over 1 to 1,000,000 passed 9,506 assertion groups with no failures. Exact counts came out as V(100) = 38, V(200) = 72, V(350) = 116 and V(700) = 213. They are exact because φ(n) is at least √(n/2), so every preimage of a value up to 700 lies below 980,000. A seed case also checked: φ(11985) = 5888 with nothing smaller, so the least preimage is more than twice the value.
The same data answers a question about keys. If U is uniform on 1 to 1,000,000, φ(U) has 163,861 distinct outputs, and its most frequent output, 241,920, occurs 937 times. Input entropy is 19.93 bits but min-entropy of the output (a measure of how guessable its most likely value is) is 10.06.
5. Prior art is related, and a negative search proves nothing
The search found Erdős and Hall (1976), Ford's Annals paper (1999), Alekseyev's inverse-totient algorithms (2014), three patents on RSA-style uses, and a 19 August 2026 preprint on structured totient preimages that treats the same kind of collision analysis as a cryptographic proposal. Under the project's rule, a negative search is not proof of non-use.
6. Source inspected, general theorem unverified
The verdict reads: source inspected, PDF page audit pending, general theorem unverified, not fully reviewed.
On the Lean side (Lean is a proof assistant that checks proofs by machine), the solution file has proof terms and no literal sorry, Lean's placeholder for a missing proof, while the comparator file has four. A clean build and dependency audit were not done.
Classes: RSA arithmetic is established use (different mathematics); preimage collisions are related precedent; a theorem-specific primitive has no documented use found; an output-entropy audit is a potentially novel application, very low confidence; the finite identities are independently validated, finite only. Passive expiry is unsupported, because copied deterministic state defeats it.
Sources and notes (11)
- Ford, The distribution of totients — the order-of-magnitude result the manuscript sharpens
- Erdos and Hall, Distinct values of Euler's phi-function (1976) — earlier fixed-dilation question
- Ford, The number of solutions of phi(x)=m (Annals 1999) — prior work on totient fibres
- Alekseyev, Computing the inverses of Euler's totient — inverse-totient algorithms already exist
- Lizama-Perez (2026), The Structured Totient Preimage Problem — preprint on the same kind of collision analysis
- Patent US20040234074A1 — RSA key-generation constraints
- Patent US5991415A — totient-based exponent blinding
- Patent US7457408B2 — modular inverse hardware
- FWD-024A source — the notes this was written from
- FWD-024A prior art — the notes this was written from
- FWD-024A applications — the notes this was written from
Cryptography researchAll 10 parts →
