Linear algebra becomes more flexible when scalars belong to a ring rather than a field. That change preserves addition and distributivity but removes division by arbitrary nonzero scalars. It explains both the power of modules and the traps in treating them as vector spaces.
Motivation
An abelian group already supports multiplication by integers: for a positive integer , define by adding to itself times, set , and use additive inverses for negative integers. The familiar identities for repeated addition are precisely the scalar laws of a module over . Studying modules therefore includes studying abelian groups, rather than merely adding an exotic generalization of vector spaces.
There is another equally useful source of examples. Suppose is a linear operator on a vector space over a field . A polynomial can act on by substituting for its variable. Asking which subspaces are stable under becomes the question of which subsets are submodules over . A single language now connects arithmetic in abelian groups with invariant subspaces and canonical forms.
These examples suggest what a definition must retain. Vectors must still add as an abelian group. Scalars must distribute over addition in both positions. The multiplication of scalars must agree with successive actions. A multiplicative identity should act as the identity. Nothing in these requirements supplies a reciprocal for every nonzero scalar, and we must not introduce that property silently in a proof.
Definitions
Throughout, is an associative ring with identity , and a module means a unital left -module unless stated otherwise. Multiplication in need not be commutative.
Definition 1.1
(A left module)
An -module consists of an abelian group and an action satisfying, for and ,
The additive identities imply and : in each case, expand an expression equal to itself plus the expression and cancel in the abelian group. They also imply . Cancellation here concerns addition in ; it says nothing about canceling multiplication by a scalar.
Definition 1.2
(Submodules, generated submodules and linear maps)
A submodule is an additive subgroup closed under every scalar in . An equivalent test is that is nonempty and for all and . To see the equivalence, use to obtain zero, then to obtain subtraction. Scalar closure follows by taking the second coefficient to be zero.
The submodule generated by consists of finite sums . It includes the empty sum, which is zero. We write it . The word finite is essential even when is infinite; there is no operation of infinite addition in an arbitrary module. A generating set need not be a basis, because relations between its elements may be unavoidable.
An -linear map preserves addition and scalar multiplication. Its kernel is and its image is . Both are submodules. The domain of a kernel and the codomain containing an image should be specified: they are not interchangeable subsets.
For a submodule , define when . This is an equivalence relation because is a subgroup. Its classes form . We propose
Before using these formulas, we must prove they do not depend on representatives. Quotient notation is a claim about a construction, not a permission to skip that proof.
Theorem/proposition
Theorem 1.3
(Quotients and their universal property)
For every submodule , the proposed operations make an -module. The projection , , is surjective and has kernel . An -linear map factors uniquely as with linear if and only if .
Theorem 1.4
(First isomorphism theorem)
For an -linear map , there is an isomorphism
No finiteness, commutativity of , or freeness assumption is required.
The first theorem describes exactly when information can descend to a quotient. The second uses the smallest amount of information that a map necessarily loses: elements differing by a kernel element have the same image. It does not identify the quotient with the entire codomain unless is surjective.
Proof sketch or proof idea
Proof
(Descent to the quotient and the induced isomorphism)
Suppose and . The difference between their proposed sums is . The difference between the scalar multiples is . Thus both operations are well defined. All module laws descend from the corresponding laws on representatives. The zero class is , and exactly when .
If , every maps to , so is necessary. Conversely, under that inclusion define . Representatives differing by have images differing by , establishing well-definedness. Additivity and scalar preservation follow from the formulas. Surjectivity of gives uniqueness: every class has a representative, so its image is forced.
Apply this construction with and regard as a map into its image. The resulting map is surjective by the definition of image. If its value at is zero, then , so its kernel is zero. An additive map with zero kernel is injective: equality of two images gives zero image for their difference. We have proved a bijective linear map. Its inverse is linear because the equations expressing preservation can be transported across a bijection. This finishes the isomorphism theorem rather than assuming that every bijection is automatically a module isomorphism.
Proof X-ray
A quotient formula has two separate obligations
The formula on representatives must first define a function on classes. Only then can injectivity or surjectivity be tested. Kernel containment supplies the first obligation; equality with the kernel supplies injectivity. Reversing that order risks proving properties of a rule that is not a function. No basis argument repairs a failed representative check.
An important extension is the submodule correspondence. Submodules with correspond to submodules of by . The inverse sends a submodule to . Images under are submodules, as are inverse images. If has the same image as some , then , so ; this proves . Surjectivity of proves the other composite is the identity. The containment hypothesis is exactly what made the first composite work.
Worked examples
Worked example 1.5
(An abelian group with unavoidable relations)
Regard as a -module. The class generates it, but is a nontrivial relation. The class generates . Define by reduction modulo two. It is well defined because changing an integer by a multiple of six does not change its parity. Its kernel is exactly , and it is surjective. Consequently . This quotient has two elements although has three; the quotient counts classes, not members of the collapsed submodule.
Worked example 1.6
(A linear operator as a polynomial action)
Let and . Define . Since powers of a single operator commute and polynomial substitution respects products, this is an -module action. Here . The subspace is an -submodule because it is -invariant. The subspace is an -subspace but is not an -submodule: leaves it. In fact generates over , and the map has kernel . Thus as an -module, while its dimension as an -vector space is two.
Worked example 1.7
(Left ideals are the correct submodules of a noncommutative ring)
Take as a left module over itself. A submodule is exactly a left ideal. Matrices whose second column is zero form a left ideal: multiplying on the left preserves a zero second column. Matrices whose second row is zero do not in general form a left ideal, since left multiplication can mix rows. The difference matters before forming a quotient. Also, left-module endomorphisms of have the form with . Under composition this identifies with the opposite ring , because composing right multiplications reverses the order of the two multiplying elements.
The last calculation also explains why the side of a scalar action is mathematical data. For a left module over a noncommutative ring, multiplying every element by a fixed scalar need not define an -linear endomorphism. Indeed, the proposed map would require for all . Such a condition holds for central scalars, but it is not a consequence of the module axioms.
What an annihilator records
For one element in a left module, its annihilator is a left ideal: sums of scalars killing still kill , and if , then . It need not be a right ideal over a noncommutative ring, because involves a different module element. In contrast, the annihilator of the whole module, consisting of scalars killing every element, is a two-sided ideal. Both products kill every element, by the two expressions just considered.
The module is called faithful when its whole-module annihilator is zero. Faithfulness says that no nonzero scalar acts as the zero operator on the entire module. It does not assert that every nonzero scalar acts injectively. For example is faithful as a -module because its first summand detects every nonzero integer, but multiplication by two kills the nonzero element in its second summand. The distinction is between killing all elements at once and killing some nonzero element. Keeping that quantifier visible is useful when passing from module actions to faithful matrix representations.
Cyclic modules over commutative rings
For this part only, let be a commutative ring with identity and a unital -module. The earlier noncommutative discussion remains more general. For , write and .
Quotienting removes exactly the ambiguity
Consider the map defined by . It is surjective by the definition of , and it preserves addition and scalar multiplication. Its kernel is the annihilator. Instead of merely invoking an isomorphism theorem, we construct the induced map and check what makes it work.
Theorem 1.8
(Cyclic-module quotient theorem)
For , the map
is an isomorphism of -modules. Here has scalar multiplication .
Proof
(Well-definedness, linearity and bijectivity)
First, is an ideal: ; if , then ; and if and , then . Hence the indicated scalar multiplication on is well-defined: if , then . Its module laws follow from the ring laws.
Now check that is well-defined. If , then , hence . Therefore : different representatives of one coset give the same output.
For addition and scalar multiplication,
Thus is an -module homomorphism. Every element of is for some , so the map is surjective. If , then , which means , so , the zero coset. This proves injectivity.
Proof X-ray
The exact equivalence behind the quotient
The entire proof is organized around one equivalence:
The forward implication proves that no two distinct cosets can have the same image. The reverse implication proves that choosing a coset representative does not change the image. Quotienting by a smaller ideal would leave some ambiguity; quotienting by a larger ideal would identify scalars that can produce different module elements.
The statement is an isomorphism of modules. An arbitrary module has no multiplication of its elements, so the notation does not automatically describe a ring. There is no need to invent a product on to use the theorem.
A submodule of a free module need not be free
Definition 1.9
(Basis and free module)
A basis of an -module is a subset such that every module element has a unique expression as a finite linear combination of its elements. A module with a basis is free.
In particular, the ring , regarded as a module over itself, is free with basis . Over a field every subspace has a basis, and it is tempting to expect the same of submodules of free modules.
Counterexample mode
A cyclic submodule with no basis
Take and its submodule
It is cyclic, generated by , but it is not free over . Suppose it had a basis . If were empty, its only finite linear combination would be zero, whereas has two elements. If were nonempty, choose . Linear independence would make the map , , injective: equality would force . This would inject a four-element set into a two-element set, which is impossible.
The obstruction is visible in the generator: although in . Indeed , and the cyclic-module theorem gives . The repaired statement is the vector-space one: over a field every subspace of a free module has a basis, while over a general ring freeness of a submodule must be checked directly.
This example uses -module structure throughout. As an additive group, is isomorphic to , but changing the scalar ring changes what “free” means: regarded as a -module, the same group has annihilator rather than , and the two ideals live in different coefficient rings. Cyclicity only guarantees that every element has at least one coefficient expression; a basis also requires that expression to be unique, and the annihilator measures the difference.
Common mistakes
Common mistake
(Dividing by a nonzero scalar)
Do not divide a module equation by a nonzero scalar. In , the equation does not force . Even over the domain , a module can have torsion. Conversely, failure of scalar cancellation does not mean additive cancellation has failed; the additive structure remains an abelian group.
Common mistake
(Quotient modules versus quotient rings)
A quotient module is generally not a quotient ring. For acting on itself on the left, a left ideal suffices for a quotient left module. A quotient ring multiplication additionally needs a two-sided ideal. Similarly, a quotient of a vector space by a subspace is a vector space, while a quotient of a module may have relations preventing any basis.
Common mistake
(Representatives are not a submodule)
Do not confuse a set of coset representatives with a submodule. In the parity example one may choose and as representatives in , but that two-element set is not closed under addition. The quotient operation is defined on classes; representatives are a calculation aid.
Summary
Modules retain additive linearity while allowing scalar rings without general division. Submodules are precisely the subsets that can be collapsed compatibly with the action. The quotient universal property tests whether a map survives that collapse. The first isomorphism theorem identifies the surviving information with the image, and its proof works without bases. Polynomial actions and integer actions will provide concrete tests for the more abstract constructions in the following chapters.
Exercises
Quick check
(Prove that every abelian group has a unique unital -module structure compatible with its addition.)
Use distributivity and to force the action of each positive integer.
Solution
The action of a positive integer is forced by distributivity and ; zero and negative integers are then forced by additive inverses. Repeated addition satisfies the four module laws by associativity and commutativity of group addition. This proves both existence and uniqueness. No divisibility of the group is assumed.
Quick check
(Let . Construct an isomorphism , including a proof that is a submodule.)
Restrict the quotient map to .
Solution
Sums of members of and are closed under addition, negatives, and scalar multiplication, so is a submodule. Restrict the quotient map to . It is surjective because . Its kernel is . The first isomorphism theorem supplies the claimed isomorphism, explicitly .
Quick check
(For , determine the kernel and image of and identify .)
Solve first.
Solution
The congruence is equivalent to . Thus the kernel is and the image is . The quotient has three classes and is isomorphic to . An explicit isomorphism to the image sends the class of to ; this class generates the image even though would be another possible generator.
Quick check
(For the operator in the second example, determine all -submodules of .)
A submodule is a -invariant subspace; test each line.
Solution
An -submodule is an -subspace invariant under . A nonzero proper subspace is a line . If , then would belong to that line, forcing its generator to have second coordinate zero, a contradiction. If , the line is and is invariant. The complete list is therefore , , and , over every field .