Skip to content

Erdős 257: settled support families, and the structure of the value space - #5044

Open
wcook04 wants to merge 10 commits into
google-deepmind:mainfrom
wcook04:fc-257
Open

Erdős 257: settled support families, and the structure of the value space#5044
wcook04 wants to merge 10 commits into
google-deepmind:mainfrom
wcook04:fc-257

Conversation

@wcook04

@wcook04 wcook04 commented Aug 18, 2026

Copy link
Copy Markdown

Closes #5054.

This is the single Erdős 257 PR. It supersedes #5033, #5040 and #5047, which I have closed — every declaration from all three is preserved here verbatim, with its own proof link.

They were separate PRs by mistake on my part. 257.lean is one file that each contribution appends to, so four PRs against it conflict by construction rather than by accident; splitting them created work for a reviewer and none of it was mine to hand over.

erdos_257 itself is untouched and remains open.

What the file now contains, in order

The open question, then what is settled about it:

Declaration Says
eventually_periodic_support eventually periodic infinite supports give irrational sums
pairwise_coprime_support so do infinite pairwise-coprime supports with summable reciprocals
factorial_support so does the positive-factorial support
two_pow_support so does the powers-of-two support

These four are stated in the same shape erdos_257 uses — ∑' n : A, 1 / (b ^ n - 1) — so they need no new vocabulary, and the base is generalised from 2 to any integer b ≥ 2.

The classical identities, tsum_top_eq and tsum_top, the latter now carrying the formal proof from #5033.

Structural results:

Declaration Says
finite_period_noncollapse for a finite support, the multiplicative order of the base modulo the reduced denominator is exactly the lcm of the support
support_measure_dichotomy the support-restricted achievement set has measure 2^(-|F|) for finite omitted F, and 0 when infinitely many are omitted
supported_achievement_set_geometry that set is also injectively coded, compact, totally disconnected and nowhere dense — and perfect when J is infinite

The last two share the six definitions in FormalConjecturesForMathlib/NumberTheory/MersenneAchievementSet.lean, the only support file this PR adds.

The geometry statement is given for arbitrary J rather than only the full set. IsTotallyDisconnected unfolds to a condition quantified over subsets, so it descends along exactly as nowhere density does; at J = Set.univ it therefore is the full-set statement, with unique coding as its injectivity conjunct. IsClosed is deliberately not a conjunct — the set is a compact subset of and Perfect already carries closedness.

Prior art

None of the mathematics here is new, and no claim of novelty or priority is made for any of it.

  • Eventually periodic supports are the $0$–$1$ coefficient case of Luca–Tachiya, doi:10.1142/S1793042113501121.
  • Pairwise-coprime supports are Erdős, On the Irrationality of Certain Series, Math. Student 36 (1968), 222–226, theorem on p. 222.
  • The factorial and powers-of-two families sit in the rapid-growth framework of Erdős–Straus, J. Indian Math. Soc. 27 (1964), 129–133; the powers-of-two case is their Example 1. Duverney later proved that value transcendental (doi:10.1017/S0305004100004783); what is recorded here is irrationality.
  • Kovač–Tao (doi:10.1007/s10474-025-01528-0, Remark 4.1) prove the strict-tail inequality for these exact weights and record both the unique coding and the Cantor-set conclusion; the achievement-set topology is classical, going back to Kakeya. The measure computation is the part their remark does not carry.

Receipts

  • lake build 'FormalConjectures.ErdosProblems.«257»' succeeds on this branch.
  • The linked adapter is compiled by CI in the source repository, not only locally: https://github.com/wcook04/plectis-lean-erdos249-257/actions/runs/32196665233 (Build foreign-corpus variant candidateslake build FormalConjecturesVariants, green on main).
  • Every linked declaration audits to [propext, Classical.choice, Quot.sound]. There is no sorryAx.
  • Each upstream statement was diffed textually against the adapter declaration its permalink points at; all match. This matters because the upstream bodies are sorry, so nothing else enforces that the link proves the stated proposition.
  • Negative control: replacing Nat.factorial (k + 1) by Nat.factorial k is rejected with a type mismatch, so the index convention is load-bearing rather than incidental.
  • Proofs are hosted externally per CONTRIBUTING.md and linked at immutable commits.

AI Usage Disclosure: the linked Lean development and this contribution were produced with AI assistance. I have reviewed the statement identity of every variant against its proof, the axiom reports, the attributions above, and this diff, and I take responsibility for the submission.

Note

These are solved variants and settled support families around Erdős 257. They do not prove irrationality for every infinite support, so erdos_257 remains open and is untouched.

One definition and one research-solved statement, with the proof linked
externally per CONTRIBUTING. Coprimality is kept: it is not a one-line
Mathlib fact.
… subseries

For a set J of allowed positive-exponent coordinates, the achievement set of
the Mersenne weights 1/(2^n - 1) restricted to J has measure 2^-|F| when the
forbidden coordinates form a finite set F, and measure 0 when infinitely many
are forbidden.

Taking J = univ gives measure exactly 1 for the unrestricted set, so the
measure-one statement is this theorem's F = empty case rather than a separate
contribution.

Six support definitions. The base-two Lambert weights and their Cantor
structure are Kovac-Tao; the measure of a fast achievement set is classical.
No novelty claim.

This does not decide whether every infinite-support subseries is irrational;
erdos_257 remains open.
@github-actions github-actions Bot added erdos-problems Erdős Problems for-mathlib touching our `FormalConjecturesForMathlib` dir labels Aug 18, 2026
@wcook04 wcook04 changed the title Erdős 257: classify the measure of support-restricted Mersenne achievement sets Erdős 257: classify the measure and the geometry of support-restricted Mersenne subseries Aug 18, 2026
@wcook04
wcook04 force-pushed the fc-257 branch 2 times, most recently from 2c5e082 to 4f8eeae Compare August 18, 2026 23:12
erdos_257 asks whether every infinite support gives an irrational sum. These
four say yes for four named families -- eventually periodic, infinite pairwise
coprime with summable reciprocals, the positive factorials, and the powers of
two -- placed directly after the open question they map.

They are stated in the same shape the open statement uses, a subtype sum over
A, so they need no new upstream vocabulary and the diff is 257.lean only. The
base is generalised from 2 to any integer b >= 2, which is what the underlying
proofs give.

None of this is new mathematics: the families are Luca-Tachiya's 0-1
coefficient case, Erdos (1968), and two cases of the Erdos-Straus rapid-growth
framework, and the docstrings say so.
A companion to the measure dichotomy over the same six definitions: for any set
J of allowed coordinates the restricted digit map is injective and its range is
compact, totally disconnected and nowhere dense, and perfect when J is infinite.

Stated for arbitrary J rather than only the full set. Total disconnectedness and
nowhere density both descend along subset inclusion, so the restricted case
costs nothing over the full one -- and at J = univ this theorem IS the full-set
statement, with unique coding as its injectivity conjunct, so a separate
full-set offer would be redundant rather than merely overlapping.

IsClosed is deliberately not a conjunct: the set is a compact subset of R and
Perfect already carries closedness.
@wcook04 wcook04 changed the title Erdős 257: classify the measure and the geometry of support-restricted Mersenne subseries Erdős 257: settled support families, and the structure of the value space Aug 19, 2026
CONTRIBUTING asks for the current year on new files.
Every other file in the repository has exactly one module docstring; the
ModuleDocstringLinter warns on the second one, and the convention elsewhere
for section headers is a plain /- ### ... -/ comment.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

erdos-problems Erdős Problems for-mathlib touching our `FormalConjecturesForMathlib` dir

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Erdős 257: add settled support families and the structure of the value space

1 participant