Ext vanishing detects freeness — under a countability hypothesis
For a countable torsion-free abelian group A, vanishing of Ext1(A, ℤ) forces A to be free. This is the Stein–Serre theorem, and it is a genuinely useful criterion: it converts a structural question into a homological one. The countability hypothesis is not decoration. Dropping it gives the Whitehead problem, which Shelah showed to be independent of the usual axioms of set theory — a rare intrusion of foundational issues into ordinary algebra.
Learning objectives
- State the Stein–Serre theorem precisely.
- Explain why torsion-freeness and countability are both needed.
- Relate the statement to the Whitehead problem.
- Use the criterion in a computation.
Section 01The statement
Let A be a countable torsion-free abelian group. Then
The converse is immediate, since free implies projective implies Ext1 = 0. So for countable torsion-free groups, freeness and Ext-vanishing are equivalent.
A torsion group can never be free, yet its Ext against ℤ is generally non-zero, so the hypothesis rules out an uninteresting failure. More importantly, the proof builds a basis by a chain argument that requires the absence of torsion at every stage.
Section 02The role of countability
The proof proceeds by writing A as an increasing union of finitely generated subgroups and constructing splittings compatibly along the chain. Countability makes that chain a sequence, so the constructions can be made one at a time.
- 1951SteinEstablished the criterion for countable torsion-free groups.
- 1950sSerreThe result appears in the homological literature and enters the standard treatments of Ext.
- 1950sWhitehead's questionDoes Ext1(A, ℤ) = 0 imply A free without the countability hypothesis?
- 1974ShelahThe Whitehead problem is independent of ZFC: it is provable under the axiom of constructibility and refutable under Martin's axiom with the negation of the continuum hypothesis.
Shelah's result means no ordinary algebraic argument can settle the uncountable case, in either direction. It is worth knowing not for its applications but because it establishes that a natural-looking homological question can be genuinely undecidable.
Section 03Using the criterion
Detecting non-freeness
ℚ is countable and torsion-free but not free, so Ext1(ℚ, ℤ) must be non-zero — and it is, uncountably so.
Subgroups of products
The Baer–Specker group ℤℕ is torsion-free and uncountable and is not free; the criterion does not apply, and separate arguments are needed.
Finitely generated case
For finitely generated torsion-free abelian groups freeness is automatic, so the criterion adds nothing — its content is entirely in the countably infinite case.
ReferenceFrequently asked questions
Does the theorem hold over a general PID?
Analogues exist for countably generated torsion-free modules over a PID, with the same shape of proof. The set-theoretic difficulties in the uncountable case persist, so the countability hypothesis is not an artefact of working over ℤ.
Is Ext<sup>1</sup>(ℚ, ℤ) computable?
It is isomorphic to the quotient of the adeles by the rationals in one description, and is an uncountable divisible torsion-free group — a ℚ-vector space of continuum dimension. Its size is what makes the failure of freeness for ℚ so decisive.
Why does this appear in a homological algebra course?
Because it is the cleanest example of Ext answering a purely structural question, and because it marks the limit of what homological methods decide. It is a useful corrective to the impression that Ext-vanishing always translates into a clean structural statement.
NavigateContinue in this stream
Curated next steps from this page. The site also surfaces algorithmically related reading below.
ProvenanceSources and further reading
This page is an original KEVOS explanatory article. It presents the underlying mathematics — definitions, algorithms, complexity results and selection criteria — in KEVOS editorial voice. No text is reproduced from any copyrighted source. Where numerical tables are relevant, KEVOS links to live authoritative databases rather than republishing static values.