A vector space is an abelian group on which the elements of a field act by scalar multiplication. If we let the scalars come from a ring instead, we obtain a module. Some of linear algebra survives this change and much of it does not; the first section of this chapter explores both. The second section specializes to the ring . Every abelian group carries a multiplication by integers: for let ( terms), let , and let for . This makes a module over , and the classification of finitely generated abelian groups becomes a problem of linear algebra over .
10.1.1 Definitions and Examples
Definition 10.1.1 (Module). Let be a ring (Definition 4.1.1). An -module is an abelian group together with a scalar multiplication , , such that for all and :
- ;
- ;
- ;
- .
As for vector spaces, and follow from the axioms: gives , and then . When is not commutative, a module in this sense is also called a left -module, since the scalars act on the left. From now on, denotes a commutative ring.
Example 10.1.2 (Vector spaces). If is a field, a -module is exactly a vector space over (Definition 4.5.1). Modules generalize vector spaces by allowing the scalars to come from a ring.
Example 10.1.3 (The modules ). The set , with componentwise addition and , is an -module. For this says that is a module over itself, the scalar multiplication being the multiplication of .
Example 10.1.4 (Abelian groups as -modules). Every abelian group is a -module, with defined as in the introduction. The axioms are the rules , , and , which follow from the definition by induction on and (the first uses that is abelian). Conversely, a -module structure on an abelian group is forced: by axioms (ii) and (iv), for , and . So a -module is the same thing as an abelian group.
Example 10.1.5 (Restriction of scalars). If is a ring homomorphism, then any -module becomes an -module via . This is called restriction of scalars. For instance, every -module is a vector space over , by restriction along the inclusion .
Example 10.1.6 (Modules over ). Let be a vector space over and a linear operator. Then becomes a -module via ; the axioms hold because and . Conversely, a -module is a vector space over (by restriction of scalars) together with the operator . So a -module is the same thing as a vector space with an operator, and different operators give different module structures on the same vector space.
10.1.2 Submodules, Ideals, and Quotients
Definition 10.1.7 (Submodule). A submodule of an -module is a subset that is an -module under the operations of . Equivalently, is a subgroup of with for all and .
Definition 10.1.8 (Ideal). An ideal of is a submodule of , viewed as a module over itself: a subgroup of such that for all and . For , the set is an ideal, the principal ideal generated by .
Example 10.1.9 (Ideals of and ). The ideals of are the subgroups with : every subgroup of has this form (Theorem 2.4.5), and each is closed under multiplication by integers. The ideals of are and the sets with . Indeed, if is an ideal, let be a nonzero polynomial of least degree; for , Euclidean division gives with or , and since , minimality forces . So .
In both rings every ideal is principal. An integral domain (a commutative ring with and no zero divisors) in which every ideal is principal is called a principal ideal domain, or PID; so and are PIDs. In both cases the reason is the Euclidean algorithm, which shows that the ideal generated by two elements is the principal ideal generated by .
If is a submodule of , the quotient group (every subgroup of an abelian group is normal) is an -module with . This is well defined: if , then . The axioms for follow from those for .
Example 10.1.10 (Quotient modules). The -module is the cyclic group . The -module is a copy of : the class of is the class of the constant , and acts as , since . We denote this module simply by .
Note. If is an ideal, the quotient is also a ring, with . This is well defined: if and , then .
10.1.3 Module Homomorphisms
Definition 10.1.11 (Module homomorphism). Let and be -modules. A map is a homomorphism of -modules (or is -linear) if for all and . The set of these maps is denoted . An isomorphism is a bijective homomorphism.
The kernel and the image of a homomorphism are submodules; for instance, if then . The First Isomorphism Theorem holds for modules: the map , , is an isomorphism of groups by Theorem 3.3.3, and it is -linear, so it is an isomorphism of -modules.
When , every homomorphism of abelian groups is -linear: for by additivity, and . So abelian groups and -modules are the same objects with the same maps; in the language of Chapter 8, the two categories coincide.
Note. Since is commutative, is itself an -module, with and . (The map is -linear because , where ; this is where commutativity is used.) For the modules and , things work as for vector spaces: a homomorphism is determined by the images of the standard vectors , and these can be arbitrary, so is isomorphic to the module of matrices with entries in (the -th column lists the entries of ). For other modules, can be zero even when and are both nonzero.
Example 10.1.12 (A vanishing ). Let and let be the module on which acts as . Then . Indeed, if is -linear, then , and a polynomial with is zero; so , and then for every .
10.1.4 Spanning Sets, Independence, and Bases
For a set , let be the module of families of elements of with for all but finitely many , with componentwise operations; for this is . A family of elements of an -module determines the homomorphism where the sum is finite. The basic notions of linear algebra can all be phrased in terms of .
Definition 10.1.13 (Spanning sets, independence, and bases). Let be a family of elements of an -module , and as above.
- The family spans (or generates ) if is surjective: every element of is a finite linear combination with . The module is finitely generated if it is spanned by a finite family.
- The family is linearly independent if is injective: implies for all .
- The family is a basis of if is bijective. A module that has a basis is free.
The span of a family (the image of ) is a submodule, written or .
If is a basis, then is an isomorphism, so the free modules are the modules isomorphic to some . The standard vectors form a basis of ; in particular is free with basis .
When , the number of elements of a finite basis is determined by the module. Indeed, an isomorphism maps onto (because ), so it induces an isomorphism . Reduction mod identifies these quotients with and , which have and elements, so . This number is the rank of the free -module. (Over a field the rank is the dimension. The rank of a free module with a finite basis is well defined over every nonzero commutative ring, but we will only need .)
Note. Over a field, every finitely generated module is free, since a finite spanning set contains a basis (Chapter 4). Over , the finitely generated free modules are the groups . The module is not free for , as we show in the next subsection.
Definition 10.1.14 (Direct sum of modules). The direct sum of -modules is the set of pairs with componentwise operations. More generally, for a family of modules , the direct sum consists of the families with and only finitely many nonzero. For example, .
The direct sum has the following universal property: homomorphisms correspond to families of homomorphisms . A homomorphism out of the direct sum restricts to each summand, and conversely a family gives the homomorphism , a finite sum. For finitely many summands, is the same module as the product , and for abelian groups we use both notations.
10.1.5 General Facts about Modules: Nothing Is True
Over a field, every vector space has a basis, every linearly independent set extends to a basis, every spanning set contains a basis, and every subspace of a finite-dimensional space is finite-dimensional. Over a general ring, all four statements fail.
Example 10.1.15 (Bases need not exist). The -module , with , has no basis. If is a nonempty family in , then is not injective: it sends to . So no nonempty family is linearly independent, and the empty family spans only .
Example 10.1.16 (Linearly independent sets need not extend to a basis). In , the element is linearly independent: implies . But no basis of contains . A basis cannot contain two elements and , because would be a nontrivial relation (basis elements are nonzero, since ). So a basis is a single element with , and the only bases of are and .
Example 10.1.17 (Spanning sets need not contain a basis). In , the set spans: , so for every . But is not linearly independent (), and neither nor spans . So this spanning set contains no basis.
Example 10.1.18 (Submodules of finitely generated modules need not be finitely generated). Let be the ring of polynomials in infinitely many variables, and . Then is generated by , but the ideal of polynomials with zero constant term is not finitely generated. Indeed, finitely many involve only finitely many variables, say . Setting is a ring homomorphism that kills each , hence kills every combination ; but it does not kill . So do not span .
Over this cannot happen: we will see that every submodule of is finitely generated (Lemma 10.2.2). More generally, it cannot happen over Noetherian rings, those in which every ideal is finitely generated, such as ; we will not prove this.
10.1.6 Exact Sequences
Definition 10.1.19 (Exact sequences of modules). A sequence of homomorphisms of -modules is exact at if , and exact if it is exact at every term that has a map in and a map out, as for groups (Definition 3.5.1). A short exact sequence is exact at , , and : is injective, is surjective, and , so that by the First Isomorphism Theorem. It splits if there is a homomorphism with .
Proposition 10.1.20 (Split exact sequences). Let be a short exact sequence of -modules.
- If satisfies , then , , is an isomorphism.
- If is free, the sequence splits; hence .
Proof. (a) By Corollary 3.5.9, applied to the underlying abelian groups, the map is an isomorphism of groups; it is -linear because and are.
(b) Let be a basis of . Since is surjective, choose with . Since , , is an isomorphism, there is a homomorphism with , namely with . Then and agree on the basis, hence everywhere. Part (a) gives . ∎