The first isomorphism theorem identifies a module after collapsing a kernel. A short exact sequence remembers that kernel and quotient together. It asks a subtler question: can the original module be rebuilt simply by putting those two pieces side by side? For vector spaces the answer is yes. For modules the answer carries information, and its failure is often the point of the subject.
Motivation
Consider reduction modulo two, . Its kernel is . Set-theoretically, every integer is either even or odd, so choosing representatives zero and one seems to give a way back from the quotient. But the representative of the sum of two odd classes is zero, whereas adding the chosen representatives gives two. Choosing representatives does not produce an additive map.
Compare a linear surjection of vector spaces . Choosing a basis of , choosing one preimage of each basis vector, and extending linearly does produce a section. The extension works because basis vectors have no relations. In , the generator has the relation , and a chosen lift must satisfy it. No odd integer is killed by two. This is the obstruction we want to express precisely.
Free modules isolate the role of basis elements. Presentations then separate arbitrary generators from the relations they satisfy. Exact sequences provide the bookkeeping that lets us recognize a kernel, a quotient, and an attempted splitting without losing track of which maps are involved.
Definitions
We work with unital left modules over an associative ring with identity. For an index set , the direct sum consists of tuples with only finitely many nonzero coordinates. Addition and scalar actions are coordinatewise. The direct product allows arbitrary tuples. They coincide for finite , but not generally for infinite .
Definition 2.1
(Free module and basis)
A module is free with basis if every element has a unique finite expression . Equivalently, is isomorphic to . Uniqueness includes the assertion that a finite relation forces every coefficient to vanish.
A generating family without uniqueness need not be a basis. Generators that satisfy relations are recorded by a second free module.
Definition 2.2
(Presentation)
A presentation of is an exact sequence . The images of the standard vectors in are generators; the image of the first map records all relations. The presentation is finite when both index sets can be chosen finite.
A finitely generated module need not be finitely presented over an arbitrary ring. The bookkeeping of kernels and images is done with exact sequences.
Definition 2.3
(Exactness, sections and retractions)
For maps , exactness at means . The sequence
is short exact if is injective, is surjective, and exactness holds at . A section of is a linear map with . A retraction of is a linear map with .
These are identities of maps, so order matters. The equation says that selects preimages compatibly with the module structure. It does not say that is injective, nor that . Indeed, will normally be a projection with kernel .
Theorem/proposition
Theorem 2.4
(Universal property of a free module)
If has basis , any choice of elements extends uniquely to an -linear map sending to . Consequently every module is a quotient of a free module.
Theorem 2.5
(Splitting criterion)
For a short exact sequence as above, the following are equivalent: has a section; has a retraction; and there is an isomorphism that identifies with inclusion of the first summand and with projection onto the second. In this situation the short exact sequence splits.
The two theorems are connected by one lifting property.
Definition 2.6
(Projective module)
A module is projective if every map lifts through every surjection : there is with .
Proposition 2.7
(Projective modules and splitting)
Every free module is projective. Every short exact sequence with projective quotient splits. A module is projective if and only if it is a direct summand of a free module.
The splitting statement concerns the quotient, and says nothing similar merely from projectivity of the kernel.
Proof sketch or proof idea
Proof
(Extending values from a basis)
For the free universal property, define . Both sums are finite, and the uniqueness of the expression in makes the definition unambiguous. Distributivity proves linearity. Every linear map with the prescribed values must satisfy the formula, giving uniqueness.
For an arbitrary module , let be its underlying set and use a formal basis vector for every . The universal property gives a map with . This map is surjective because is already an image. Its kernel is the module of relations. Generating that kernel by another free module supplies a presentation, although the sets of generators and relations may be very large.
Proof
(Section, retraction and direct sum)
Now assume a section is given. Define
For , the element lies in . Injectivity of gives a unique with . Thus , proving surjectivity. If , apply to obtain and then use injectivity of to obtain . Therefore is an isomorphism. Projection onto the first coordinate of its inverse gives a retraction.
Conversely, suppose . For , choose with and propose . If is another lift, then for some . The proposed values differ by . Hence the definition is independent of the choice. It is linear, as follows by choosing sums and scalar multiples of lifts, and . This proves a section exists. Finally the direct-sum description immediately supplies both a section and a retraction. All three conditions are therefore equivalent.
The proof is short, but each hypothesis of a short exact sequence is used at a different point.
Proof X-ray
Where exactness and injectivity enter the splitting
Surjectivity of is the step that needs : the correction is killed by because , and exactness turns that into membership of . Injectivity of is then used twice, once to make the coordinate unique and once to finish after has removed the -coordinate. The converse has one non-obvious move: correct an arbitrary lift by subtracting . Two lifts differ by an element of , and on the map vanishes because ; that is exactly why the corrected value is independent of the lift. Surjectivity of is what guarantees that a lift exists at all.
The last block of the proof turns lifting through a free module into the projective characterization.
Proof
(Free modules, projectivity and direct summands)
Free modules are projective: lift the image of every basis vector through the given surjection and extend by the universal property. For an infinite basis, this assertion uses the relevant choice of preimages; in the usual set-theoretic foundations with the axiom of choice, it holds for all free modules. Finite bases require only finitely many choices.
A useful characterization is that is projective if and only if it is a direct summand of a free module. For the forward direction, take a free surjection and lift ; splitting gives . Conversely, suppose is free, with projection . Given and a surjection , the map lifts to with because is projective. Restricting to gives . This proves the lifting condition rather than treating “direct summand” as a synonym without justification. Finally, a short exact sequence with projective quotient splits: lifting through gives a section.
Worked examples
Worked example 2.8
(A short exact sequence that does not split)
The sequence , where and is reduction modulo two, is exact. If a section existed, would be an odd integer . Linearity would imply , impossible. Equivalently, cannot be isomorphic to , because the latter has a nonzero element killed by two while the former does not. The kernel is free, so this example also disproves the claim that a free kernel guarantees splitting.
Worked example 2.9
(A projective module that is not free)
Let and . Since in , multiplication by three is an idempotent endomorphism of the module . Its image is , and : the intersection is zero, and in . Thus is projective. It is not free. A nonzero free -module contains a copy of , which has six elements, whereas has only two. This example separates projectivity from freeness using a fully explicit decomposition.
Worked example 2.10
(Relations and a concrete splitting)
Take , , and . The kernel of is exactly the image of , and is a section. The isomorphism supplied by the proof is . For its inverse gives . A compatible retraction is . Other sections exist, for example ; a split sequence need not come with a preferred splitting.
For a finite presentation over a commutative ring, matrices turn these ideas into calculations. If the columns of a matrix describe relation vectors in , then the presented module is . Changing bases in the generator module or relation module does not change the resulting isomorphism class. Over a noncommutative ring the side conventions of matrices need attention; the structural statements above do not depend on choosing a matrix convention.
There is also a practical way to compare two splittings. If are sections, then . Hence their difference factors uniquely through the injective map , giving a map with . Conversely is a section for every such . Once a splitting exists, the freedom in choosing it is therefore described exactly by .
Mapping out of a presentation
A finite presentation also tells us how to construct a map without guessing its values on every element. If , choose proposed images for its generators in a target module . They extend to a map from by freeness. The quotient universal property says that this map descends to exactly when every relation column maps to zero. Thus a lift from a presented module requires compatible preimages of the generators, rather than arbitrary preimages. This gives a concrete way to locate the obstruction in a nonsplit sequence: the quotient relations hold after applying the surjection, but their chosen lifts may produce nonzero kernel elements. Merely knowing that such errors lie in the kernel does not make them vanish. A section must correct all of them simultaneously.
Concept lensAlgebraic
A section as correction of all relation errors
In a presented quotient, arbitrary lifts of generators usually satisfy each relation only up to an element of the kernel. Changing the lifts changes those errors. A section is a simultaneous choice for which every error vanishes. In the parity sequence, replacing the lift of the odd class by any other odd integer never makes its double zero. This turns the abstract splitting obstruction into a specific unsatisfiable relation.
Common mistakes
Common mistake
(A zero composite is not exactness)
Exactness is a local condition on two consecutive maps. It is stronger than the assertion that their composite is zero: that assertion gives only . Equality must be checked. The zero maps between nonzero modules give an immediate warning that a zero composite alone proves very little.
An abstract isomorphism is not, by itself, the definition of a splitting of particular maps. The isomorphism must be compatible with and ; the explicit map in the proof is valuable because its compatibility is visible.
Counterexample mode
An abstract isomorphism with the direct sum does not split the sequence
Claim: if as modules, then splits. Let and consider
with and , reducing the first coordinate modulo two. It is exact. Both and are direct sums of countably infinitely many copies of and of , so reindexing gives . Yet the sequence does not split. If , write with and . Since , we get , while is the nonzero element of , contradicting . The correct criterion asks for an isomorphism compatible with and , as in the splitting theorem.
The same caution applies to infinite families of coordinate vectors.
Common mistake
(Coordinate vectors do not span a direct product)
For infinite families, the standard coordinate vectors generate the direct sum but generally not the direct product. For example the tuple lies in but is no finite sum of standard coordinate vectors. This observation does not by itself settle whether an arbitrary product is abstractly free; it only rules out that proposed basis.
Summary
Free modules let us extend arbitrary values on basis vectors. Presentations record the relations that prevent the same extension on arbitrary generating sets. A short exact sequence splits precisely when a linear section, a linear retraction, or a compatible direct-sum decomposition exists. Free quotients split sequences, free kernels need not, and projective modules capture the exact lifting property behind the argument.
Exercises
Quick check
(Let be reduction modulo three. Prove that no -linear section exists, then identify exactly which step of the free-module lifting argument fails.)
Apply a proposed section to the relation .
Solution · Answer
A section sends to an integer congruent to one modulo three. Since , linearity requires , hence , a contradiction. The quotient is generated by but is not free on that generator. An arbitrary lift of the generator need not satisfy its relation, so the free universal property is unavailable.
Quick check
(Show that the short exact sequence , with and , splits. Give a section.)
Look for an element of that is congruent to one modulo three and killed by three.
Solution · Answer
Define in . It is well defined because changing by three changes by twelve. It is additive and satisfies since four is congruent to one modulo three. Exactness follows from . The resulting direct sum is the familiar decomposition of the cyclic group of order six.
Quick check
(If satisfies , prove as left modules, without assuming is central.)
Multiply an element of the intersection on the right by .
Solution · Answer
Every equals . If is in the intersection, multiply on the right by : the first expression gives , while the second gives . Thus the intersection is zero. Both pieces are left ideals, so this is a decomposition of left modules. Centrality would matter for extra ring-product claims, which are not needed here.
Quick check
(Prove that a finite direct sum of projective modules is projective directly from the lifting definition.)
Lift summand by summand.
Solution · Answer
Given and a surjection , restrict to each summand and choose a lift using projectivity. Define . This is linear and . The construction proves exactly the lifting property of the finite direct sum.