What does a quotient forget?
A short exact sequence records a subgroup and a quotient, but those two groups alone need not determine the middle group. For example, both and have a subgroup and quotient isomorphic to . The difference is in how the two pieces are joined.
All groups in this unit are abelian and written additively, equivalently they are -modules. Fix an abelian group and an integer . Consider a short exact sequence
Exactness says that is injective, is surjective, and .
Definition 9.1.1
(Splitting and equivalence with fixed endpoints)
- A splitting of the sequence is a homomorphism with .
- Two such sequences are equivalent with fixed endpoints when an isomorphism of their middle groups commutes with the two maps and acts as the identity on the specified copies of and .
If a splitting exists, , , is an isomorphism: it is surjective by subtracting a lift of the quotient, and its kernel is zero by applying and then using injectivity of .
For equivalence, merely knowing that the middle groups are isomorphic is weaker; it does not identify the embedded subgroup and quotient map.
An invariant from a lift
Choose with . Because , there is a unique with . This depends on the chosen lift: every other lift has the form , which changes to . Therefore the class in is independent of the lift. An equivalence fixing endpoints sends lifts to lifts and preserves this class.
Common mistake
(Reading the invariant in instead of )
The element itself is not an invariant. For , the lift gives , while the lift gives , so . Only the class is independent of the lift.
Theorem 9.1.2
(Cyclic extension classification)
The assignment above is a bijection between equivalence classes of short exact sequences with fixed endpoints and elements of . The sequence splits exactly when its assigned class is zero.
Proof
(A presentation by one generator and one relation)
For construct
The map is well defined since the relation changes by a multiple of ; it is surjective. If , then for some integer . Since , in forces and ; hence is injective. If , write . The relation gives , so the kernel is exactly . The element maps to and satisfies . Every class in is therefore attained.
Conversely, given an extension and a lift with , the map , , is onto: the quotient is generated by . Its kernel is exactly . Indeed, a kernel element has after applying , and then injectivity of gives . Thus it induces an equivalence with the endpoints fixed.
Finally, if , the map on sends to . It and its inverse descend to an equivalence . Hence equal classes give equivalent extensions; the lift calculation shows different classes cannot. If , choose a lift with ; sending to defines a splitting. Conversely a splitting supplies a lift of killed by , so its class is zero. ∎
The proof has three separate tasks, and each uses a different hypothesis.
Proof X-ray
Realize, recognize, and change the lift
Realization builds from one generator and the single relation ; injectivity of is where enters, since must force . Recognition shows that every extension is some : the map is onto because the cyclic quotient is generated by , and its kernel is computed by applying and then using injectivity of . Independence of the lift is the shear , which fixes the subgroup and the quotient while moving by . The splitting criterion is then the special case , where a lift with exists.
The proof also covers : both the quotient and are zero, so there is one split class. No finiteness or divisibility assumption on was used.
Worked examples
Worked example 9.1.3
(Two ways to join two groups of order two)
Take and . Since , the invariant has two possible values. The zero class gives the split middle group . The nonzero class is realized by , with and . A lift has , so the class is nonzero and the sequence cannot split.
Worked example 9.1.4
(The same middle group can carry different extensions)
Let and . The classes and in are different, hence and are not equivalent with fixed endpoints. Both middle groups are nevertheless cyclic of order : exactness gives , and has order five, so has order when . An abstract isomorphism of middle groups need not respect both endpoint maps.
Worked example 9.1.5
(Coprime orders force splitting)
Take and . Multiplication by is bijective on , so and every extension splits. For instance with and is exact: is well defined and injective with image . The lift satisfies , so is a splitting and .
A splitting has to respect the group law, which is exactly what the next example tests.
Counterexample mode
A section as a set map is not enough
For , choosing representatives , gives a set-theoretic section. It is not a homomorphism: in the quotient, but in the middle group. A splitting must preserve the group operation.
Exercises
Quick check
(For and , how many fixed-endpoint equivalence classes are there? Give a representative for the zero class.)
Compute .
Solution · Answer
There are three classes, corresponding to . The zero class is represented by using the standard inclusion and projection.
Quick check
(In the same setting, show that represents the class and does not split.)
Use the lift .
Solution · Answer
The middle element maps to in the quotient and , so its invariant is . It is nonzero, hence the sequence is nonsplit. Equivalently, no element of mapping to has order dividing three.
Quick check
(Suppose multiplication by is surjective on . What can be said about every such extension?)
Compute .
Solution · Answer
Surjectivity gives , so has one element. The theorem says every extension splits; this conclusion does not require a preferred splitting.
Quick check
(Explain why an equivalence of sequences must preserve the invariant in .)
Follow a lift of through the equivalence.
Solution · Answer
An equivalence sends a lift of to a lift of and fixes . Thus it preserves the unique element satisfying for those corresponding lifts, and hence its class. Other choices of lift change only by .
Further reading
The Stacks Project, Extensions defines extensions and morphisms with fixed endpoints in the broader setting of abelian categories. The elementary classification here is proved directly for abelian groups.