Lecture 29: Free modules, finite generation, and bases

for vector spaces

1. Universal property of free modules

Recall:

Definition 29.1. Let R be a ring. Then the direct sum module

Rn := R ⊕ . . . ⊕ R
is called the free R-module of rank n.

Chit-chat 29.2. Why is this called a free R-module? Behold:

Proposition 29.3. Let M be an R-module. Then any ordered n-tuple of

elements x1 , . . . , xn ∈ M uniquely determines an R-module homomorphism
X : Rn → M
given by
(0, . . . , 0, 1, 0, . . . , 0) �→ xi
where the 1 is in the ith coordinate.

Remark 29.4. This is the same property as for the free group on n genera-
tors: Any ordered n-tuple of elements of a group G determines a unique map
from Fn to G.

Proof. Given (x1 , . . . , xn ), define X : Rn → M by

X(a1 , . . . , an ) := a1 x1 + . . . + an xn ∈ M.
This is a group homomorphism because
X((a1 , . . . , an ) + (b1 , . . . , bn )) = (a1 + b1 )x1 + . . . (an + bn )xn
= (a1 x1 + . . . + an xn ) + (b1 x1 + . . . bn xn )
= X(a1 , . . . , an ) + X(b1 , . . . , bn ).
37
LECTURE
38 29: FREE MODULES, FINITE GENERATION, AND BASES FOR VECTOR SPACES

where the middle equality is using the property of M being an R-module. This
is also an R-module homomorphism because
X(r(a1 , . . . , an )) = X((ra1 , . . . , ran ))
= (ra1 )x1 + . . . + (ran )xn
= r(a1 x1 + . . . an xn )
= rX((a1 , . . . , an ))
Again, the penultimate equality is using the fact that M is an R-module. �

2. Spans and linear independence and bases

Definition 29.5. Fix x1 , . . . , xn ∈ M .
(1) We say this collection spans M if the map X : Rn → M is a surjection.
(2) We say that this collection is linearly independent in M if the map
X : Rn → M is an injection.
(3) We say this collection is a basis for M if X is both an injection and a
surjection.

Chit-chat 29.6. You’ll recognize these terms from linear algebra. And in
terms of equations, these definitions mean exactly what you’d imagine:

Proposition 29.7. Let M be a left R-module, and let x1 , . . . , xn ∈ M be

an ordered collection.
(1) The collection spans M if and only if for every y ∈ M , there exists a
collection a1 , . . . , an ∈ R so that
y = a1 x1 + . . . + an xn .
(2) The collection is linearly independent if and only if the equation
0 = a1 x 1 + . . . + an x n
has one and only one solution: (a1 , . . . , an ) = (0, . . . , 0).
(3) The collection is a basis if and only if for any y ∈ M , the equation
y = a1 x 1 + . . . + an x n
has one and only one collection (a1 , . . . , an ) solving it.
Proof. The first is the definition of surjection. The latter claim follows
because a homomorphism is injective if and only if the kernel is trivial, and
(0, . . . , 0) ∈ Rn is the additive identity of Rn . The last claim is the definition
of a bijection. �

Definition 29.8. We say that a module is finitely generated if there is some

number n ∈ Z≥0 and a surjective R-module homomoprhism Rn → M .
 4. SPANNING SETS ARE BIGGER THAN INDEPENDENT SETS 39

Chit-chat 29.9. This is also in analogy to groups. A group G is finitely

generated if and only if there is some finite collection of elements gi such
that all other elements can be expressed as products of gi and their inverses.
Likewise, M is finitely generated if there is a finite collection xi such that every
element of M can be obtained by taking linear combinations of xi .

Non-example 29.10. Not every module over R admits a basis. This is in

contrast to vector spaces. For example, if R = Z and M = Z/nZ, then for
any x ∈ M , the equation
ax = 0
has many solutions—a could equal n, 2n, . . ..
Take-away: Not every finitely generated R-module admits a basis.

3. Vector spaces and subspaces

Recall:

Definition 29.11. A commutative ring is called a field if R − {0} is a group

under multiplication.

Definition 29.12. Let F be a field. A module over F is called a vector space

over F .

Definition 29.13. Let V be a vector space. Then a submodule of V is called

a linear subspace of V .

4. Spanning sets are bigger than independent sets

The following is the most importance consequences of being a field, as
opposed to a ring:

Theorem 29.14. Let F be a field, and let M be a vector space over F . If

v1 , . . . , vn span and w1 , . . . , wm are linearly independent, then n ≥ m.
Proof of the Theorem. Let y1 , . . . , ym be linearly independent, and
let v1 , . . . , vn be spanning. By re-ordering vi if necessary, we can assume that
y 1 = a1 v 1 + . . . + an v n
for a1 �= 0. Then y1 , v2 , . . . , vn is also spanning, for we can obtain v1 as a linear
combination of the y1 and the vi —just divide the above equation by a1 �= 0
and rearrange terms.
Let M1 ⊂ M be the submodule generated by y1 —i.e., the image of R → M
defined by 1 �→ y1 —and consider the quotient
M/M1 .
LECTURE
40 29: FREE MODULES, FINITE GENERATION, AND BASES FOR VECTOR SPACES

(You’ll prove this is also an R-module—i.e., a vector space—in your home-

work.) Then y2 , . . . , ym are still linearly independent, for a linear combination
of them equals zero if and only if
a1 y1 = a2 y2 + . . . + am ym
for some a1 ∈ F , and such an equation can hold only when all the ai = 0,
since the yi are assumed linearly independent. Note y1 = 0, v2 , . . . , vn are still
spanning, so v2 , . . . , vn is spanning. So we have n − 1 vectors spanning M/M1 ,
and we have m − 1 linearly independent vectors in it.
By repeating the trick above, if we have m linearly independent elements in
a vector space spanned by n elements, we can obtain m−k linearly independent
elements in a quotient vector space spanned by n − k elements. So which of
these numbers will hit 0 first? If n − k = 0 first, we are in a quotient vector
space spanned by 0 elements—i.e., the zero vector space—so we must conclude
m − k = 0 as well, for there are no linearly independent vectors in the zero
vector space. And in this case, m = n. If m − k reaches zero before n − k
does, we have that m ≤ n. �

5. Corollaries
Corollary 29.15. If M is a finitely generated vector space, then any two
bases of M have the same number of elements in it.

Proof. If the {vi } and {wi } above are both spanning and linearly inde-
pendent, we have n ≥ m and m ≥ n. Hence m = n. �

Definition 29.16. Let M be a finitely generated vector space over F . Then

the number of elements in a basis for M is called the dimension of M over F .

Remark 29.17. This is the single most important fact in linear algebra:
That we have a notion of dimension. It took us thousands of years to know
what we mean by an n-dimensional space, so don’t take this lightly!

Corollary 29.18. If M is a finitely generated vector space, any linearly

independent collection w1 , . . . , wm can be completed to a basis—that is, we can
find wm+1 , . . . , wn so that the resulting collection w1 , . . . , wn is both linearly
independent and spanning.

Proof. Since M is finitely generated, there is some N for which we have a

surjection F N → M . So any set of linearly independent vectors must have size
≤ N by the theorem. If Xm : F m → M is the map determined by w1 , . . . , wm ,
and if Xm is not surjective, choose an element wm+1 not in im(Xm ). Note the
 5. COROLLARIES 41

resulting collection w1 , . . . , wm+1 is still linearly independent, for if

a1 w1 + . . . + am+1 wm+1 = 0
then we have
a1 w1 + . . . + am wm = −am+1 wm+1 .
If am+1 = 0, by linear independence of the wi , we know all ai = 0. On the
other hand, if am+1 �= 0 we arrive at a contradiction by dividing:
a1 am
w1 + . . . + wm = wm+1 .
−am+1 −am+1
The lefthand side is in the image of Xm , but wm+1 was chosen not to be.
So we have an injective homomorphism Xm+1 : F m+1 → M . If Xm+1 is
not surjective, we repeat the argument. It must become a surjective map at
some m + k ≤ N by the theorem. So let k be the integer at which Xm+k first
becomes a surjection. By the above argument, it is still an injection, so we
have a basis determined by the generators w1 , . . . , wm+k . �

Corollary 29.19. Any finitely generated module over a field F is isomorphic

to F n for some n.
Proof. Begin with the linearly independent set 0 and complete to a basis.
A basis defines an isomorphism from F n to your module. �

Remark 29.20. This is definitely not true for R-modules if R is not a field—
after all, any finite abelian group is a Z-module, but any free Z-module is the
zero module or an infinite module.

Corollary 29.21. Let V � ⊂ V be a subspace. Then dim V � + dim V /V � =

dim V .
Proof. Let v1 , . . . , vdim V � be a basis for V � . Let u1 , . . . , udim V /V � be a bsis
for V /V � . Then choosing representatives ui for ui , the set
v1 , . . . , vdim V � , u1 , . . . , udim V /V �
is a basis for V . It obviously spans since for each a ∈ V , a is a linear combi-
nation of ui , hence a is in the V � -orbit of some linear combination of the ui .
It is linearly independent because if we have that
0 = a1 v1 + . . . adim V � vdim V � + b1 u1 + . . . + bdim V /V � udim V /V �
then
0 = a1 v 1 + . . . adim V � vdim V � + b1 u1 + . . . bdim V /V � udim V /V � .
The ai terms go to zero since v i = 0, hence we get an equation saying a linear
combination of the ui is zero. This means each bi must be zero by linear
LECTURE
42 29: FREE MODULES, FINITE GENERATION, AND BASES FOR VECTOR SPACES
�
independence of the ui . The original equation then says that 0 = ai vi , so
by linear independence of the vi , the ai must be zero. �

Corollary 29.22 (Rank-nullity theorem). Let f : V → W be a map of

F -modules and assume V is finitely generated. Then dim ker f + dim im f =
dim V .
Proof. By the first isomorphism theorem, we know there is a group iso-
morphism V / ker f ∼
= im f . But this homomorphism is also an F -module map,
as you can check by hand. Thus im f ∼ = V / ker f . �