Modules

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Finitely generated modules

Shameek Paul

Let M be a finitely generated R-module. Let {x1 , . . . , xk } be a spanning


set for M over R. We have an onto R-linear map Rk → M where we send
(a1 , . . . , ak ) to a1 x1 + · · · + ak xk . So by the first isomorphism theorem, we get
that M ' Rk /ker. Hence, any finitely generated R-module is isomorphic to a
quotient of Rk for some k. When k = 1, we say that the module is cyclic. The
zero module is a cyclic module.
For x ∈ M , let Ann x = {a ∈ R : ax = 0}. Observe that if {x} is a
generating set for a cyclic module M , then the kernel of the above map R → M
is precisely Ann x. We now prove the splitting property of free modules. This
says that if f : M → N is an R-linear map whose image is a free module, then
M is isomorphic to the product of ker f and image f . An equivalent statement
for modules having a finite basis will be, if we have an onto map M → Rk , then
M ' ker f × Rk .
Let M be an R-module and f : M → Rk be an onto R-linear map. Let
e1 , . . . , ek denote the standard basis of the free module Rk . As f is onto, for
each i there exists xi ∈ M such that f (xi ) = ei . Define g : Rk → M by
g(a1 , . . . , ak ) = a1 x1 + · · · + ak xk . Then g is an R-linear map. For each i, we
have f ◦ g(ei ) = ei and so f ◦ g is the identity map on Rk . Hence, g is injective
and so g gives an isomorphism from Rk to g(Rk ).
We claim that M is the internal direct sum of ker f and g(Rk ). From our
claim, it will follow that M ' ker f × Rk . Let us prove the claim. Let x ∈ M .
To get an element in g(Rk ), we will first go to Rk by using f and then check
that x − g(f (x)) is in ker f . So if x ∈ M , we have x ∈ ker f + g(Rk ). Hence,
M = ker f + g(Rk ). Suppose x ∈ ker f ∩ g(Rk ). Then f (x) = 0 and there
exists y ∈ Rk such that g(y) = x. So y = f ◦ g(y) = 0 and hence x = 0. This
proves our claim.
Let M be an R-module. An element x ∈ M is called a torsion element if
there exists a ∈ R \{0} such that ax = 0. Let T or(M ) = {x ∈ M : x is a torsion
element}. Observe that 0 ∈ T or(M ). We say that M is a torsion module if
T or(M ) = M . We say that M is torsion-free if T or(M ) = {0}. Observe that a

1
ring R is an integral domain if and only if it is a torsion-free R-module.
Let R be an integral domain. If M is a free R-module, then M is torsion free.
The converse is not true in general. The module Q is a torsion-free Z-module.
More generally, if R is a subring of an integral domain S, then S is a torsion-free
R-module. However, Q is not free.
This is because any two rational numbers satisfy a Z-linear relation and so
they are linearly dependent over Z. Hence, if the Z-module Q had a basis, then
it must have exactly one element. But if this happens, we get the contradiction
that Q is a cyclic group.
The module Q is not a finitely generated Z-module. Can we get a finitely
generated Z-module which is torsion-free but not free? We will see later that
this is not possible. This will follow from the theorem that a finitely generated
module over a PID which is torsion free will be a free module.
When R is an integral domain, T or(M ) is a submodule of M . Let M 0 denote
the quotient module M/T or(M ). Then M 0 is a torsion free R-module.
Let R be a PID. If M is a finitely generated R-module, then M 0 = M/T or(M )
is a finitely generated torsion-free R-module and so by the above theorem, M 0
is a free R-module. As the quotient map M → M 0 is onto and its kernel is
T or(M ), by the splitting property of free modules (this is true over any ring),
we get that M ' T or(M ) × M 0 .
Later, we will also prove the theorem that for any commutative ring R,
if an R-module has a spanning set of k elements and an independent set of l
elements, then l ≤ k. From this result (just as we have seen in vector spaces), it
will follow that any two bases of a free R-module must have the same number
of elements. We will call this number the rank of the free module.
So, we get the following corollary. If R is a PID, a finitely generated R-
module M is isomorphic to T or(M ) × M 0 , where M 0 is a free R-module. In
addition, if M has a generating set of k elements, then the rank of M 0 is at
most k.
We have seen the theorem that if R is a PID, then any submodule of Rk is
a free R-module and has rank at most k. Let us recall the proof where we use
induction on k.
When k = 0, then Rk = {0} and so the only submodule is {0}. Let k > 0 and
N be a submodule of Rk . Consider the map p : N → R given by p(a1 , . . . , ak ) =
a1 . The kernel of the map p is a submodule of {0} × Rk−1 which is isomorphic
to Rk−1 . So by the induction hypothesis, the submodule ker p is a free module
having rank at most k − 1.
As R is a PID, the image of the map p is also a free module and it has

2
rank at most 1. From the splitting property of free modules, we get that N is
isomorphic to the product of ker p and image p. As both are free modules, it
follows that N is a free module having rank at most k.
We will now proceed to prove a more refined version of the above corollary
for modules over Euclidean domains. We will then use this result for the rings
Z and F [X], where F is any field. Let M be a finitely generated module over a
Euclidean domain R.
We have seen that any finitely generated module is isomorphic to a quotient
of Rk for some k. So it is enough to study R-modules of the form Rk /N where
N is some submodule of Rk . As R is a ED and hence a PID, we know that N
is a free module having rank at most k. Let {y1 , . . . , yl } be a basis of N with
l ≤ k.
We want to change the bases of Rk and N to get bases {x1 , . . . , xk } of Rk
and {z1 , . . . , zl } of N such that for 1 ≤ i ≤ l, there exist di ∈ R such that
zi = di xi . We will then say that the two bases are compatible. If we can find
compatible bases, then we claim that Rk /N ' R/d1 R × . . . × R/dl R × Rk−l .
We will first prove this claim and then show the procedure to find compatible
bases.
Let {x1 , . . . , xk } and {z1 , . . . , zl } be compatible bases of Rk and N . For
l + 1 ≤ i ≤ k, let di = 0. Define a map Rk → R/d1 R × . . . × R/dk R by sending
a1 x1 + . . . + ak xk to ( [a1 ], . . . , [ak ] ). This is an onto R-linear map. Suppose
a1 x1 + . . . + ak xk is in the kernel. Then for 1 ≤ i ≤ k, we have [ai ] = [0] and so
ai ∈ di R.
So for 1 ≤ i ≤ k, there exist bi ∈ R such that a1 x1 +· · ·+ak xk = b1 d1 x1 +. . .+
bk dk xk . This is an R-linear combination of d1 x1 , . . . , dk xk . As for l + 1 ≤ i ≤ k,
we have di = 0, this is in the R-span of z1 , . . . , zl . Thus the kernel is contained in
N . As for 1 ≤ i ≤ l, we can check that zi is in the kernel, so N is also contained
in the kernel. By the first isomorphism theorem, we see that our claim is true.
Recall that an R-linear map f : Rn → Rm corresponds to an m × n matrix
whose j th column is f (ej ). As N has a basis of l elements, the inclusion N ⊆ Rk
gives an R-linear map Rl → Rk . Let A be the k × l matrix which corresponds
to this map.
If the bases for Rk and N are compatible, then this matrix has two blocks.
The upper block is a diagonal matrix (d1 , . . . , dl ) and the lower block is a zero
matrix of size k − l × l. We will say that such a matrix is in compatible form.
Conversely, if the corresponding matrix is in compatible form, then the two
bases must be compatible.
We have seen that row operations correspond to changing the basis of Rk and

3
the column operations correspond to changing the basis of N . So it is enough to
show that we can bring any k × l matrix into compatible form by using row and
column operations. It is only at this point that we use the division algorithm.
Till now we have only used that the Euclidean domain R is a PID.
Recall that an ED has a valuation function v : R \ {0} → Z≥0 . We will call
the valuation of a non-zero element its size. For example, the size of an integer
is its modulus and the size of a polynomial in F [X] is its degree.
Given a matrix A, we look for a smallest (w.r.t. size) non-zero element and
bring it to the (1, 1) position by interchanging rows and columns. If all the
elements in the first row are multiples of a11 , then we use column operations to
make them zero.
If some element in the first row is not a multiple of a11 , then we use a column
operation to replace it with an element (remainder) which is smaller than a11 .
We then bring it to the (1, 1) position. This process will terminate in a finite
number of steps as the sizes of the elements are bounded below by zero.
Once we reach the stage where all the elements in the first row are zero
except the element in the (1, 1) position, we see if there are any elements in the
first column which are not divisible by a11 . If there is any such element, we
replace it by a smaller element by a row operation. Then we bring the smaller
element to the (1, 1) position and start checking the first row again.
After finitely many steps, we will reach the stage where all the elements
in the first row as well as in the first column are zero except the element in
the (1, 1) position. We now check if there are any elements in the lower-right
(k − 1) × (l − 1) submatrix A0 which are not divisible by a11 .
Eventually we will reach the stage where all the elements in the first row as
well in the first column are zero except a11 and that all the elements in A0 are
multiples of a11 . We now let d1 be the element in the (1, 1) position.
By the induction hypothesis, we can bring A0 into compatible form by row
and column operations. These operations can be performed on the correspond-
ing rows and columns of A without changing the zeroes in the first row and first
column of A. So we are able to bring A into compatible form.
Observe that the gcd of the elements of a matrix which is in compatible form
is d1 and that when we perform row or column operations, the gcd of elements
of the new matrix is the same as that of the old matrix. So we see that d1 is
the gcd of the elements of A and so it is only determined upto a unit.
By the induction hypothesis, we see that the compatible form for A0 is also
unique upto units. Let the diagonal matrix in the compatible form for A0 be
(d2 , . . . , al ). As d1 divides all the elements of A0 and as d2 is the gcd of the

4
elements of A0 , so d1 |d2 . So we see that the elements in the diagonal block of
the compatible form of A also satisfy d1 | d2 | . . . | dl .
Observe that if d ∈ R, then R/dR is a cyclic R-module. Also any cyclic
R-module is isomorphic to a quotient of R by an ideal (which is the annihilator
of a generator). As R is a PID, so there exists d ∈ R such that this ideal
is dR. Thus we have arrived at the structure theorem of finitely generated
modules over Euclidean domains. This says that any such module is isomorphic
to a finite product of cyclic R-modules. More specifically, it is isomorphic to
R/d1 R × . . . × R/dl R × Rk−l where d1 | d2 | . . . | dl .
As {z1 , . . . , zl } is a basis of N and as for 1 ≤ i ≤ l, we have zi = di xi , so none
of the di ’s can be zero. If some di is a unit, then the quotient R/di R is trivial.
When R = Z, we can further specify that the di ’s are positive to get unique
values. When R = F [X], we can specify that the di ’s are monic polynomials to
get unique values.
Suppose M = R/c1 R × . . . × R/cs R where c1 | c2 | . . . | cs and none of the
ci ’s are units. Show that a generating set of M which has the least number of
elements, must have exactly s elements. If we take any generating subset of
M which has s0 elements where s0 > s, then in the diagonal submatrix of the
compatible form, there will be s0 − s units.
If M is a finitely generated torsion module over a Euclidean domain, then
l must be equal to k. Finite abelian groups are finitely generated torsion Z-
modules. From the above theorem, we see that the converse is also true as
any finitely generated torsion Z-module must be a finite product of finite cyclic
groups and so it must be a finite abelian group.
In particular, if G is a finite abelian group, then we see that G is isomorphic
to Z/n1 Z × . . . Z/nr Z, where n1 | n2 | . . . | nr and the ni ’s are positive. If r is
the smallest size of a generating set, then none of the ni ’s are 1.
Let V be a vector space over a field F and dimk V = n. Let T : V → V
be a linear map. Then V is a F [X]-module via T . If m(X) is the minimal
polynomial of T , then for any v ∈ V , we have m(X).v = m(T )v = 0. So V is
a torsion F [X]-module. As V is a finitely generated F -module, it is a finitely
generated F [X]-module and we can find a generating set which is a subset of
the basis.
As F is a field, F [X] is a Euclidean domain and so by the above theorem,
we see that V is isomorphic to F [X]/h p1 (X) i × . . . × F [X]/h pm (X) i, where
the pi (X)’s are monic and p1 (X) | p2 (X) | . . . | pm (X). The pi (X)’s which are
non-constant are called the invariant factors of the module V . Observe that
pm (X) is the minimal polynomial of T .

5
Assignment

Obtain the Smith normal form of the following matrices over Z.


 
0 2 −1
1. −3 8 3
2 −4 −1
 
5 9 5
2. 2 4 2 
1 1 −3
Obtain the Smith normal form of the following matrix over Q[X].
 
−x − 3 2 0
3.  1 −x 1 
1 −3 −x − 2

You might also like