VibeMathedMath problems solved by AI
All problems

Erdős Problem #424

Erdős problem #424 · erdosproblems.com/424

Let a1=2a_1 = 2 and a2=3a_2 = 3 and continue the sequence by appending to a1,,ana_1, \dots, a_n all possible values of aiaj1a_ia_j - 1 with iji \ne j. Is it true that the set of integers which eventually appear has positive density?

Formal statementwhat was actually proved, as formalized
/-- The recursively generated set in Erdős Problem 424.  Using an inductivepredicate makes the phrase "the smallest set" literal: its constructors areexactly the two seeds and the closure rule. -/inductive Generated : ℕ → Prop where  | two : Generated 2  | three : Generated 3  | mul_sub_one {a b : ℕ} : Generated a → Generated b → a ≠ b →      Generated (a * b - 1) /-- Number of generated positive integers at most `x`. -/noncomputable def countUpTo (x : ℕ) : ℕ := by  classical  exact ((Finset.Icc 1 x).filter Generated).card /-- The exact conclusion called "positive lower density" in the paper. -/def HasPositiveLowerDensity : Prop :=  ∃ c : ℝ, 0 < c ∧ ∀ᶠ x : ℕ in Filter.atTop,    c * (x : ℝ) ≤ (countUpTo x : ℝ) theorem theorem_one : HasPositiveLowerDensity := …  /-- The new numbers obtained from `A` by applying `x * y - 1` to distinctmembers of `A`.  This agrees verbatim with the `formal-conjectures`definition. -/def nextGeneration (A : Set ℕ) : Set ℕ :=  {z : ℕ | ∃ x y, x ∈ A ∧ y ∈ A ∧ x ≠ y ∧ z = x * y - 1} /-- The finite-stage closure used by `formal-conjectures`. -/def sequenceSet : ℕ → Set ℕ  | 0 => {2, 3}  | n + 1 => sequenceSet n ∪ nextGeneration (sequenceSet n) /-- The union of all finite stages, as in `formal-conjectures`. -/def generatedSet : Set ℕ := ⋃ n : ℕ, sequenceSet n open Classical in/-- The intended problem statement in the staged-set formulation. -/theorem erdos_424_lower_density :    ∃ c : ℝ, 0 < c ∧ ∀ᶠ x : ℕ in Filter.atTop,      c * (x : ℝ) ≤        (((Finset.Icc 1 x).filter fun n ↦ n ∈ generatedSet).card : ℝ) :=  generatedSet_has_positive_lower_density

A machine checks that the proof entails this statement. Whether this statement is the problem is for you to judge. Source

Result
Proved(see note)
Status
Candidate (review pending)
AI contribution
AI co-developed
Method
Argument
Field
Number Theory, Integer Sequences
Posed by
Douglas Hofstadter
Year posed
1977
Years open
49y
Solved
2026-07-20
Model
GPT-5.6 Pro
Vendor
OpenAI
Collaborators
Samuel Korsky
Verification
Lean-verified
Publication
Announced
Significance
10 / 100
Disclosed cost
Wikipedia
No dedicated article

What was actually shown

Proves positive lower density. The Formal Conjectures encoding asks for Set.HasPosDensity, a density that exists and is positive; erdosproblems.com says Erdos most likely meant lower density.

What the AI did

GPT-5.6 Pro developed the argument together with Samuel Korsky, in particular searching for the transition matrices the interval-partition argument needs. The Lean formalization was produced separately, with Codex, by Boris Alexeev.

Verification

Lean 4.32.0 and Mathlib v4.32.0 formalization by Boris Alexeev, produced with Codex, from the informal argument of Samuel Korsky and GPT-5.6 Pro. Rebuilt independently on 2026-08-02 against Lean 4.32.0 and Mathlib v4.32.0 (the file as published, sha256 ca4a2371918b1a7c66dccfe324305298): all 6,394 lines compile in 1,215 s with no sorry and no admit, and #print axioms reports the top theorem depending on exactly [propext, Classical.choice, Quot.sound], the three standard Lean axioms, with no sorryAx and nothing assumed. The formalized conclusion is positive lower density, stated against the same nextGeneration, sequenceSet and generatedSet definitions the Formal Conjectures statement of #424 uses. Status is candidate rather than resolved because erdosproblems.com has not accepted the claim: its proof-claims page states plainly that appearing there is no guarantee of correctness and does not mean anyone associated with the site examined any part of the proof.

Sources

erdosproblems.com

Submitted by GoldenMongoose827

Changelog16 changes
  • Rasmus Lindahlchanged Formal statement from 28 lines (generatedSet referenced but not defined) to 43 lines (self-contained)
  • Rasmus Lindahlset Formal statement to Lean statement from plby/lean-proofs
  • Rasmus Lindahlchanged Result qualifier from Two qualifications. erdosproblems.com still lists #424 as open, and states that a filed pr… to Proves positive lower density. The Formal Conjectures encoding asks for Set.HasPosDensity,…
  • Rasmus Lindahlchanged What the AI did from GPT-5.6 performed the search for the necessary transition matrices. to GPT-5.6 Pro developed the argument together with Samuel Korsky, in particular searching fo…
  • Rasmus Lindahlchanged Model from GPT-5.6 to GPT-5.6 Pro
  • Rasmus Lindahlset More links to Lean proof (Erdos424.lean), Proof claim on erdosproblems.com, Manuscript (PDF), Formal Con…
  • Rasmus Lindahlchanged Verification note from Formalized in Lean by Boris Alexeev and Aristotle. to Lean 4.32.0 and Mathlib v4.32.0 formalization by Boris Alexeev, produced with Codex, from …
  • Rasmus Lindahlchanged Short name from Erdős Problem #424 to Erdős #424
  • Rasmus Lindahlset Field detail to Number Theory, Integer Sequences
  • Rasmus Lindahlchanged Entry address from erdos-problem-424 to erdos-424
  • Rasmus Lindahlset Publication to Announcement
  • Rasmus Lindahlapproved this entry
  • Rasmus Lindahlset Significance to 10
  • Rasmus Lindahlset Result qualifier to Two qualifications. erdosproblems.com still lists #424 as open, and states that a filed pr…
  • Rasmus Lindahlchanged Status from Resolved to Candidate (review pending)
  • GoldenMongoose827submitted this entry

Discussion