Proved theorem Lean Unrefereed

Odd perfect numbers: the equal-exponent cases through r = 167

No dated manuscript catalogued; standalone verification archive linked above. On this site since · last revised · Verify & co-author
Kind
Theorem, computer-assisted (certificates found by search, each case checked by the Lean proof assistant's kernel)
Status
Proved, Lean-checked, case by case, after two falsifier agents' grades; unrefereed
Prior work
27 of the 36 cases through r = 157 extend the case list found in the dated searches; the rest are classical, credited below
On this page
  1. The statement
  2. Check it yourself
  3. Worked examples
  4. Status and who checked it
  5. Prior work
  6. Further reading

The statement

A number is perfect when its divisors other than itself add up to it, as 6=1+2+36 = 1 + 2 + 3. Nobody knows whether an odd perfect number exists. Euler showed that one would be a power of one prime, the special prime, times other primes each raised to an even power.

For an odd prime rr, the statement S(r)S(r) is: no NN is odd and perfect with

N=παq1 r−1⋯qt r−1,N = \pi^{\alpha} q_1^{\,r-1} \cdots q_t^{\,r-1},

where π,q1,…,qt\pi,\allowbreak q_1,\allowbreak \ldots,\allowbreak q_t are distinct primes and π≡α≡1(mod4)\pi \equiv \alpha \equiv 1 \pmod 4. Writing the common exponent as 2β2\beta, so r=2β+1r = 2\beta + 1, this is the classical equal-exponent case β\beta.

Result. S(r)S(r) holds for every odd prime rr from 3 to 157 (36 cases), and for r=163r = 163 and r=167r = 167, each checked in Lean with no hypothesis left over. The exclusion holds for numbers of any size with that exponent pattern.

Check it yourself

The method on r=3r = 3, every prime but the special one squared. Write Φ(q)=1+q+⋯+qr−1\Phi(q) = 1 + q + \cdots + q^{r-1}, the sum of the divisors of qr−1q^{r-1}; every odd divisor of Φ(q)\Phi(q) must divide NN. Then 3 must divide NN, and:

  • Φ(3)=1+3+9=13\Phi(3) = 1 + 3 + 9 = 13, so 13 divides NN;
  • Φ(13)=1+13+169=183=3×61\Phi(13) = 1 + 13 + 169 = 183 = 3 \times 61, so 61 divides NN;
  • Φ(61)=3783=3×13×97\Phi(61) = 3783 = 3 \times 13 \times 97, so 97 divides NN;
  • Φ(97)=9507=3×3169\Phi(97) = 9507 = 3 \times 3169 brings 3169.

That is four primes (13, 61, 97, 3169) that leave remainder 1 on division by 3. At most one is special, so three of them each put a 3 into the divisor sum of NN; but NN holds only 323^2. Finishing the case takes 13, 61 and 97 as the special prime in turn. A certificate is the written record of such a chain for one rr, checkable line by line in whole-number arithmetic.

The fact that rr is not the special prime, for r=5r = 5: take q=11q = 11, which leaves remainder 1 on division by 5. Then 1+11+121+1331+14641=16105=5×32211 + 11 + 121 + 1331 + 14641 = 16105 = 5 \times 3221: the factor 5 appears once, and 3221=5×644+13221 = 5 \times 644 + 1. Take q=3q = 3, which does not: 1+3+9+27+81=121=5×24+11 + 3 + 9 + 27 + 81 = 121 = 5 \times 24 + 1.

Which cases were open. The printed lists exclude β\beta = 1, 2, 3, 5, 6, 8, 11, 12, 14, 17, 18, 24, 62, every β\beta that leaves remainder 1 on division by 3, and every β\beta that leaves remainder 2 on division by 5. Strike those from 15 to 36 and ten remain: 15, 20, 21, 23, 26, 29, 30, 33, 35, 36. For each, 2β+12\beta + 1 is a prime: 31, 41, 43, 47, 53, 59, 61, 67, 71, 73.

The kernel's verdict, the last lines of the Lean check for r=163r = 163: "Lean default kernel accepts the solution" and "Your solution is okay!", for all 9 theorems of that run.

Worked examples

  • r=7r = 7, so β=3\beta = 3. Hagis and McDaniel settled β=3\beta = 3 in 1972, so this case is a known result checked again by another method. Since 7 leaves remainder 3 on division by 4, the method proves by itself that 7 is not the special prime.
  • r=89r = 89, so β=44\beta = 44. The exponent is 88, and 89=4×22+189 = 4 \times 22 + 1 leaves remainder 1 on division by 4. Those are the cases that need Kanold's 1941 fact that the special prime is not rr; it is now itself proved in Lean, so no case rests on it as a hypothesis.
  • r=19r = 19, so β=9\beta = 9. The least β\beta left open by the printed lists, and one of the five that Cohen and Williams tried and could not settle.

Status and who checked it

  • Proved, Lean-checked. For each rr, a search wrote a certificate, two checking programs accepted it, and a Lean theorem turns an accepted certificate into S(r)S(r). Each run that counts rebuilt the proof from source on a fresh machine; the kernel accepted it, and the axioms used were only the three permitted. Each case's label reads "proved; not known to us in print, by the dated search".
  • Who checked it. Each label was given only after two falsifier agents graded the case with no blocking objection open: the falsifier of record and a blind second falsifier agent that never sees the other's grade. Unrefereed.
  • Open. The cases past r=167r = 167; and the general question, whether any odd perfect number exists.

Prior work

Credited. The excluded values of β\beta in print: β=1\beta = 1 (Steuerwald 1937), 2 (Kanold 1941), 3 (Hagis and McDaniel 1972), 5, 12, 17, 24 and 62 (McDaniel and Hagis 1975), 6, 8, 11, 14 and 18 (Cohen and Williams 1985), every β\beta with remainder 1 on division by 3 (McDaniel 1970) and every β\beta with remainder 2 on division by 5 (Fletcher, Nielsen and Ochem 2012, as Yamada attributes it; Brown 2015). Euler's form, and Kanold's fact that the special prime is not rr, are classical. Those cases of the 36 are known results checked again here.

What the searches found. Dated searches per batch of cases covered OpenAlex, Semantic Scholar, arXiv, zbMATH Open, Crossref, HAL, OEIS, GitHub, Zenodo, Stack Exchange, Google Scholar and full texts of the equal-exponent papers. An outside reviewer agent, reading them, counts 27 of the 36 cases through r=157r = 157 as extending the case list found in the dated searches. Five of them, β\beta = 9, 15, 20, 21 and 23, are the ones Cohen and Williams wrote they "also looked at … without further success". A 2023 paper by Gallardo and Rahavandrainy says in passing that "all cases of β less than 20 are resolved", although the list it sums up holds neither 9 nor 15. The same authors' Theorem 1.1 (printed 2011) rules out, in the cases searched, the part whose special prime is 5.

A deduction, not a theorem of the source. The same reviewer agent notes that this case list, with the classical exclusions, implies β≥38\beta \ge 38 (a common exponent of at least 76) for any odd perfect number of this shape; β=38\beta = 38 is the least value it leaves open.

Still owed. Kanold's 1941 paper and the publisher-walled papers of the 1970s were read only through later citations.

Further reading