Erdős Problem #274
Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: open (not automatically the live result).
Subject areas: group theory, covering systems.
Read this status correctly
The status above is reproduced from GCL's immutable external-catalogue snapshot. It is not a new claim of current mathematical resolution, a source-semantic review, an independent proof check, or GCL certification. A published proof and a GCL replay are separate events.
Editorial authority: The protected V2 ledger determines acceptance. This manuscript is a documentary explanation of the source record, not mathematical certification; a source return, formal statement, or external proof repository cannot substitute for a checked proof.
The question in plain language
Erdős’s 1977 formulation asks about a finite abelian group \(G\). Choose subgroups \(H_1,\ldots,H_k\) and representatives \(a_1,\ldots,a_k\in G\), with \(k>1\). Can the cosets be an exact partition of \(G\) while having different sizes? In symbols, the proposed counterexample would satisfy
and \(|a_iH_i|\ne|a_jH_j|\) whenever \(i\ne j\). The answer to the original abelian question is no: some two cosets must have the same size.
For a finite group, \(|a_iH_i|=|H_i|\) and
Thus “two cosets have the same size” is equivalent to “two subgroup indices are equal.” The broader Herzog–Schönheim conjecture replaces “finite abelian” by an arbitrary group and uses finite indices explicitly: a finite partition into cosets of proper subgroups with \(k>1\) should contain \(i\ne j\) with \([G:H_i]=[G:H_j]\). That broader statement is the modern open problem. The distinction matters: the catalogue’s single open label conflates a proved special case with an open generalization.
Its quantifiers can be written as
For an infinite \(G\), the hypothesis is still a finite partition and the indices, not the cardinalities of the cosets, are the invariant being compared.
A small example
Work additively in \(C_6=\mathbb Z/6\mathbb Z\). The following five cosets form an exact partition:
The first four parts are cosets of the subgroup \(\{0\}\) and have size \(1\); the last is the coset \(\{1,4\}\) of \(\{0,3\}\) and has size \(2\). The sizes are therefore \((1,1,1,1,2)\), illustrating the repetition forced by the conjecture. Exact counting gives
Here is an independently reproducible integer check of union, disjointness, and sizes; it uses no floating-point arithmetic.
G = set(range(6))
parts = [{0}, {2}, {3}, {5}, {1, 4}]
assert set().union(*parts) == G
assert sum(len(part) for part in parts) == len(G) == 6
assert all(parts[i].isdisjoint(parts[j])
for i in range(len(parts)) for j in range(i))
print(sorted(set().union(*parts)))
print([len(part) for part in parts])
The output is [0, 1, 2, 3, 4, 5] and [1, 1, 1, 1, 2]. This finite witness is a check of the example, not evidence that the global conjecture has been proved or disproved. The authenticated source audit records the same example and also reports an exhaustive \(C_6\) coset enumeration with no all-distinct-size exact cover.
What is known
Zhi-Wei Sun’s On the Herzog–Schönheim conjecture for uniform covers of groups, Journal of Algebra 273 (2004), 153–175, Theorem 1.1, proves that a nontrivial uniform cover whose subgroup factors are subnormal cannot have pairwise distinct finite indices. In a finite abelian group every subgroup is normal, hence subnormal, and an exact partition is a uniform cover of multiplicity one. This theorem therefore proves Erdős’s finite-abelian question. The theorem does not prove the unrestricted conjecture for arbitrary groups.
There are substantial partial results. Margolis and Schnabel, The Herzog–Schönheim conjecture for small groups and harmonic subgroups, Beiträge zur Algebra und Geometrie 60 (2019), 399–418, state Theorem A: every group of order less than \(1440\) satisfies the conjecture. Garonzi and Margolis, The Herzog–Schönheim conjecture for simple and symmetric groups, arXiv:2509.25118v2 (5 May 2026), prove the conjecture for symmetric groups (Theorem 1.2) and simple groups (Theorem 1.3). These results leave the general finite-group problem open; they are not a counterexample and not a universal proof.
The maintained Erdős entry says precisely that the abelian special case is settled by Sun while the broader question remains open. Its comments and linked proof claims are useful status evidence, but they are not themselves a mathematical certificate.
GCL's current state
The authenticated RESULT/1 source return on original issue #1544 has disposition STATUS_CONFLICT. It identifies the 1977 abelian formulation, the 1980 restatement, Sun’s theorem, the small-group and simple/symmetric-group results, and the exact \(C_6\) calculation. Its conclusion is the correct split status: proved for the original finite-abelian scope; open for the broader Herzog–Schönheim conjecture.
The pinned catalogue packet is teorth/erdosproblems commit 536408716724a2402bdfa7c1d764bf83d64a8c89, data/problems.yaml, Git blob 4e60263a80af6207aca494d2e1388e44fa372889, entry 274. It records informal status open, formal status unformalized, statement-formalized metadata yes, and last update 2025-08-31. The fixed Programme registry classifies the entry as PROTECTED_READY_BACKLOG and binds it to issue #1544; those are workflow facts, not a change to the mathematics.
The pinned FormalConjectures source is FormalConjectures/ErdosProblems/274.lean at commit 85f863718beeec7b58a3a1926ee92e3472bc2020, Git blob abb606f9afc58b385707761cc2ff1c512dd5d603. Its universal theorem and its herzog_schonheim theorem both end in by sorry. The file’s abelian declaration carries an external formal_proof link to Jostamon/erdos274-hs-abelian commit 2ab8a2e39e7dd7836adf577b52555f069244466f, but neither that repository nor its CI was replayed here. The current provider revision also changes the main statement to an existential counterexample form and retains sorry; this semantic drift is not silently treated as an equivalent theorem. A pinned Lean statement, sorry, external proof repository, or self-reporting CI is not mathematical certification.
What remains
For the original abelian problem, the remaining work is documentary and semantic: preserve the primary-source wording, explain the reduction from equal coset sizes to equal indices, and reconcile it with the formal structure. The 1974 Canadian Mathematical Bulletin Research Problem No. 9 scan was not obtained in the audit, so its exact wording is not independently asserted.
For the broad conjecture, a bounded next task would enumerate groups of order exactly \(1440\), after fixing a GAP/SmallGroups release and hash. Candidate distinct indices \(d_i\) must first pass the exact necessary equation
because the disjoint cosets have total size \(|G|\). Exact subgroup and coset enumeration could then test every candidate partition. One verified partition with pairwise distinct indices would disprove the conjecture; exhausting order \(1440\) would establish only that order. Resource exhaustion without a certificate is PARTIAL, not a negative result. The 1997 source scan also remains unavailable, and its cited printed page differs from the live entry’s reference, so that historical wording stays an explicit exception.
Sources and claim boundaries
- Original sources inspected: Erdős, Problems and results on combinatorial number theory III (1977), printed p. 49, scan; Erdős and Graham, Old and New Problems and Results in Combinatorial Number Theory (1980), printed p. 26, scan. The later 1997 source cited by the live entry, scan, was not retrievable; the audit records its p. 49 versus live p. 53 discrepancy.
- Maintained Erdős entry #274, last edited 31 October 2025: live status and bibliography, not certification.
- Pinned catalogue entry, commit and blob identified above.
- Sun, arXiv full text, Theorem 1.1; journal metadata and pages are given above.
- Margolis–Schnabel, arXiv full text, Theorem A; Garonzi–Margolis, arXiv v2, Theorems 1.2 and 1.3.
- Exact authenticated GCL source return, disposition
STATUS_CONFLICT, including the source and semantic exceptions. - Pinned FormalConjectures file, commit and blob identified above. Its declarations are statement infrastructure, not checked proofs.
The primary literature, catalogue metadata, live entry, source-audit return, formal statement, and external repository are different evidence classes. This article reports their boundaries rather than upgrading one class into another. The protected V2 ledger determines acceptance.
Catalogue and evidence provenance
- Pinned catalogue source: teorth/erdosproblems at
536408716724. - GCL issue: #1544 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
- GCL triage: PROTECTED_READY_BACKLOG.
- Formal-statement coverage: A provider statement was included in the pinned snapshot. Presence does not mean the statement or proof has been checked.
- External mathematical status: consult current problem page and its dated bibliography.