Catalog
Every problem here was open when it entered the pool. Each entry states the problem in ordinary mathematical language and gives you the exact Lean statement you would need to prove. Nothing is paraphrased, so what you read is what gets checked.
A proof outlives the network that paid for it.
Whatever becomes of this subnet, a conjecture settled here stays settled: in the record, readable, and rerunnable by anyone who doubts it.
Number theory
Date added 25 Aug 2026
Erdős problem 1137 Let d n = p n + 1 − p n d_n=p_{n+1}-p_n d n = p n + 1 − p n , where
denotes the
th prime. Is it true that
max n < x d n d n − 1 ( max n < x d n ) 2 → 0 \frac{\max_{n < x}d_{n}d_{n-1}}{(\max_{n < x}d_n)^2}\to 0 ( max n < x d n ) 2 max n < x d n d n − 1 → 0 as
?
$1,401 bounty · 1 piece from 1 person· last one 3 days ago
week of 18 Sept 2025 · nothing published week of 25 Sept 2025 · nothing published week of 2 Oct 2025 · nothing published week of 9 Oct 2025 · nothing published week of 16 Oct 2025 · nothing published week of 23 Oct 2025 · nothing published
Solve it A complete proof or refutation in Lean. The first one the validator accepts takes the whole bounty.
Add a piece A lemma, a definition, a special case. There is no race: when someone finishes the problem, the pool splits between everyone whose work led there.
Command line, then a pull request on GitHub.
← Erdős problem 1135 Erdős problem 1142 → What has been built 1 piece · 28 lemmas
8 Sept 2026
Proof target
import FormalConjectures.ErdosProblems.«1137»
import TaskSupport
namespace Bounty
theorem target : fcTypeOfName% "Erdos1137.erdos_1137" := by
sorry
end Bounty
Open this target’s Challenge.lean Original conjecture (Lean type)
True ↔
Filter.Tendsto
(fun x => ↑((Finset.range x).sup fun n => primeGap n * primeGap (n - 1)) / ↑((Finset.range x).sup primeGap) ^ 2)
Filter.atTop (nhds 0)Original conjecture source: FormalConjectures/ErdosProblems/1137.lean
Source type SHA-256 sha256:a62d844e5fab8e4377daba553a74aa8ae79e3a98cbf6f1b7c3031dbc18f6895d
Task id fc-8432eac9-erdos1137-erdos-1137-a887e1b05b-formalized-v1
Task commitment sha256:0246b46a5fc432433b67785337b6f52c94b276505c8c440d7fb5d48abc0c8838 week of 30 Oct 2025 · nothing published
week of 6 Nov 2025 · nothing published
week of 13 Nov 2025 · nothing published
week of 20 Nov 2025 · nothing published
week of 27 Nov 2025 · nothing published
week of 4 Dec 2025 · nothing published
week of 11 Dec 2025 · nothing published
week of 18 Dec 2025 · nothing published
week of 25 Dec 2025 · nothing published
week of 1 Jan 2026 · nothing published
week of 8 Jan 2026 · nothing published
week of 15 Jan 2026 · nothing published
week of 22 Jan 2026 · nothing published
week of 29 Jan 2026 · nothing published
week of 5 Feb 2026 · nothing published
week of 12 Feb 2026 · nothing published
week of 19 Feb 2026 · nothing published
week of 26 Feb 2026 · nothing published
week of 5 Mar 2026 · nothing published
week of 12 Mar 2026 · nothing published
week of 19 Mar 2026 · nothing published
week of 26 Mar 2026 · nothing published
week of 2 Apr 2026 · nothing published
week of 9 Apr 2026 · nothing published
week of 16 Apr 2026 · nothing published
week of 23 Apr 2026 · nothing published
week of 30 Apr 2026 · nothing published
week of 7 May 2026 · nothing published
week of 14 May 2026 · nothing published
week of 21 May 2026 · nothing published
week of 28 May 2026 · nothing published
week of 4 Jun 2026 · nothing published
week of 11 Jun 2026 · nothing published
week of 18 Jun 2026 · nothing published
week of 25 Jun 2026 · nothing published
week of 2 Jul 2026 · nothing published
week of 9 Jul 2026 · nothing published
week of 16 Jul 2026 · nothing published
week of 23 Jul 2026 · nothing published
week of 30 Jul 2026 · nothing published
week of 6 Aug 2026 · nothing published
week of 13 Aug 2026 · nothing published
week of 20 Aug 2026 · nothing published
week of 27 Aug 2026 · nothing published
week of 3 Sept 2026 · 1 piece published
week of 10 Sept 2026 · nothing published
18 Sept 2025 a year of work on this problem 10 Sept 2026
erdosproblems.com/1137
Something wrong with this formalization?
A statement that does not faithfully capture the original conjecture is the one real risk here, so we would rather hear about it early - before someone spends weeks on it.
Our channel is in the Bittensor Discord server. Join the server first, then open the channel to send your report.
Reward
$1,401
for closing this problem
Whoever closes it takes the bounty. Published pieces that helped get there share a separate contribution pool, paid out once the problem is closed.
Set by bounty policy dynamic-age-v2-locked: the amount is worked out from how long the problem has stood open, so it moves as the pool and the pool's age profile move.
Prove it or break it
The same problem accepts both. You can submit a proof of the statement or a counterexample that refutes it, and you choose which when you submit.
Prove the statement no attempts
Find a counterexample no attempts
Erdős problem 1137 · Conjectures.io