How large would a unique totient preimage have to be?
- Kind
- Computer-assisted lower bound
- Status
- Retained batch summaries checked; a separate common-divisor theorem has a fully replayable certificate; unrefereed
- Prior work
- Extends the forced-divisor method of Carmichael, Klee and Ford
On this page
The statement
Euler’s totient counts the integers from 1 through that are coprime to . Carmichael conjectured that every value it takes occurs at least twice. If a value occurs only once, both that value and its unique preimage must exceed .
The same argument bounds how common such values could be. Among all totient values, the lower limiting proportion with a unique preimage is less than .
Check it yourself
and both have totient 12. Such a replacement produces a second preimage wherever the same factors can be exchanged without disturbing the others. A hypothetical unique preimage has to prevent every such exchange. Following the consequences forces prime squares into it.
The checker reconstructs the seed primes and combines the retained batch summaries. Rechecking every prime in the historical computation requires regenerating the retired batch files. The Lean proof covers the smaller input bound .
A separate result with a complete certificate
If every preimage of a totient value is divisible by ten, that value is at least . The checker archive includes all 8,243 roots of the case tree, the arithmetic verifier and the structural checks that every case reaches a justified conclusion. Carmichael’s conjecture remains open.
Prior work
Searches run. Primary full texts of Carmichael (1907, 1922), Klee (1947), Erdős (1958), Pomerance (1974), Brillhart–Lehmer–Selfridge (1975), Schlafly–Wagon (1994), and Ford (1998, 2014) were checked. Dated searches of arXiv (via DataCite), zbMATH Open, OEIS, MathSciNet MR Lookup, Zenodo, CORE, and OpenAlex (including citations of Ford 1998 and Schlafly–Wagon 1994) found no counterexample lower bound above Ford's 1998 bound and no lower bound on totients whose preimages are all divisible by ten.
Still owed. Google Scholar (unavailable at its shared monthly floor), Semantic Scholar, Math Stack Exchange, MathOverflow, HAL, OSF, MathSciNet beyond MR Lookup, paywalled journals, print handbooks (Guy; Sándor and Crstici), and full texts of Ramadan-Jradi (1998) and Beasley (2017).