Posted onEdited onInExplorationWord count in article: 1.2kReading time ≈4 mins.
This article replaces vector spaces by modules and follows the
resulting path from presentations and Smith normal form to finitely
generated abelian groups and rational canonical form.
Replacing a field by a ring removes division by nonzero scalars.
Modules need not have bases, and submodules of free modules need not be
direct summands. Throughout this article is a nonzero ring with identity.
Module basics
Definition. A left -module is an abelian group with an action satisfying
A submodule is
an additive subgroup closed under the action of . The submodules of the left regular
module are exactly the left
ideals of .
Definition. An -linear map preserves addition and satisfies
. Its kernel and image
are submodules. Its cokernel is .
In particular, every -linear
map is a group homomorphism between the underlying additive groups of
its modules.
Definition. For a submodule , the quotient has
Products and direct sums of modules use componentwise operations.
Theorem.First isomorphism theorem for
modules. For an -linear
map ,
Linear operators as modules
Let be a field. If is an -vector space and , define
Conversely, restricting an -module to the constant polynomials
gives an -vector space and
multiplication by gives an -linear operator.
Proposition. The constructions above give an
equivalence between -modules
and pairs consisting of an
-vector space and an -linear endomorphism. Moreover, -linear maps are exactly the -linear maps commuting with the
operators.
Free modules and
presentations
Definition. A subset is a basis if every
has a unique finite
expression . A
module with a basis is free. If , then .
For free modules with chosen finite bases, homomorphisms are represented by matrices (acting on column
vectors). Over a noncommutative ring this convention is for left modules
and coefficients must be placed consistently.
Proposition. Every module generated by is a quotient of . More precisely, the map is
surjective, and its kernel is the module of relations among the
generators.
Finite generation does not imply freeness: is generated by one
element as a -module but
is not free. A nonzero relation in one chosen presentation does not
itself prove nonfreeness. For example,
this presentation has a nonzero relation module although its quotient
is free.
For nonzero commutative rings, any two bases of a finite free module
have the same cardinality; this number is its rank. (Arbitrary
noncommutative rings need not have this invariant-basis-number
property.)
Smith normal form
Theorem.Smith normal form. Let
be a PID and . There are
invertible matrices
and such that
where every . The nonzero
, normalized up to units, are
uniquely determined and are called the invariant factors.
Allowed reductions. Smith reduction uses the following
invertible operations, applied to either rows or columns:
interchange two rows (or two columns);
multiply a row (or column) by a unit of ;
add a multiple of one row (or column) to another.
Equivalently, these operations replace by with and . ◻
Corollary. Every subgroup of a free abelian group
is free of rank at most
.
Proof. Induct on ,
with immediate. Choose a
coordinate projection . If , apply induction in . Otherwise
for some . Put and choose with . For , write ; then , uniquely. Hence
Induction makes free of rank
at most , completing the
proof. ◻
Proposition. Every finitely generated module over a PID has a finite
presentation
equivalently, is the cokernel
of an injective map between finite free modules.
Proof. For a surjection and an
isomorphism , the
presentation is
◻
Finitely generated abelian
groups
Theorem.Structure theorem for finitely
generated abelian groups. Every finitely generated abelian
group has a decomposition
The rank and invariant factors
are unique.
Uniqueness. The free rank is intrinsic:
The torsion subgroup and
its -primary subgroup are also intrinsic. If , then for every ,
These dimensions recover the multiset of exponents for every prime . Moreover, .
Padding the exponent lists on the left with zeros to length and using uniquely recovers
each . ◻
The first summand is the free part and the finite direct sum is the
torsion part. The torsion part can alternatively be decomposed into
cyclic groups of prime-power order (the elementary-divisor form). Its
exponent is when , i.e. the least positive integer
annihilating every torsion element.
Application to linear
operators
Since is a PID, the
structure theorem applies to the -module associated with a linear
operator .
Theorem.Invariant-factor
decomposition. Let be
finite-dimensional over and let
. There
are unique monic nonconstant polynomials such that,
as -modules,
There is no free summand,
because is
infinite-dimensional over . The
companion matrices of the give
the rational canonical form of .