← All problems

Erdős Problem #690 — Let dk(p)d_k(p) be the density of those integers whose kkth smallest prime factor is pp (i.

47 years old · traced to

Let dk(p)d_k(p) be the density of those integers whose kkth smallest prime factor is pp (i.e. if p1<p2<⋯p_1<p_2<\cdots are the primes dividing nn then pk=pp_k=p). For fixed k≥1k\geq 1 is dk(p)d_k(p) unimodular in pp? That is, it first increases in pp until its maximum then decreases.

References

Progress summary

Refreshed
Open · solution reported

A new preprint claims to settle the problem completely, but its proof has not received independent mathematical verification.

Erdős asked whether, for each fixed kk, the densities indexed by the prime pp rise to one maximum and then fall. The known classification is complete for 1≤k≤201\leq k\leq 20, while the remaining cases had been open.

Known results

  • Cambie proved unimodality for 1≤k≤31\leq k\leq 3.
  • Cambie proved non-unimodality for 4≤k≤204\leq k\leq 20.
  • Cambie proved eventual decrease for every fixed kk.

Recent claimed complete resolution

Shouqiao Wang and Davide Crapis claim that the sequence is non-unimodal for every k≥4k\geq 4, using prime-gap arguments, certified computations, and a Chinese-remainder construction. Their preprint says the proof was produced by “Fields” with limited human interaction; this is an unverified claim, with no independent expert assessment found.

Community submission (unverified), October 28, 2026

A submitted paper argues that non-unimodality holds for every k≥4k\geq 4, which would settle the classification; the submission itself is unverified.

Current status (as of October 2026): The cases 1≤k≤31\leq k\leq 3 and 4≤k≤204\leq k\leq 20 are established, while the claimed result for every k≥4k\geq 4 remains unverified.

Sources

Solutions 1

Lean verifiedThis is an adaptation of the Wang-Crapis argument that solves the problem for k > 20, by simplifying the range needed for finite certificates to cover as well as removing the dependence on Dusart's unconditional theta bounds, as those are infeasible to formalize in Lean in reasonable time due to having to verify zeta zero location and completeness on the critical line for many billions of zeroes. It has been formalized completely in Lean with nothing but the 3 standard axioms.See full solutionHide full solution
  • lean-repository.json61,555 bytesGitHubOpen
    Lean verification did not pass
    Why it did not pass
    19 declaration(s) are outside `namespace Submission` (Nat.count_mul_period, Nat.count_eq_mod_add_div_mul, Nat.countSubtypeFinEquiv). Everything an answer adds goes inside it, so two answers can never collide
  • research-paper.pdf95,627 bytesOpen
  • lean-repository.json61,537 bytesGitHubOpen
  • lean-repository.json61,469 bytesGitHubOpen
    Lean verification did not pass
    Why it did not pass
    answers must not extend Lean syntax or execute elaborator code (infixr); inline local notation