CHAPTER 2.

VECTOR SPACES

In Chapter 1, we just assumed that we knew what

a vector is: it’s some sort of thing with a direction

and a length. We also focused on dimensions 2

and 3, to keep things simple. For the same rea-

son, we kept abstraction to a minimum, because we

needed to focus on other things, such as answering

the question: why should I care about linear alge-

bra? (MOTIVATION!) In other words, we looked

at linear algebra from the point of view of someone

who wants to apply it.

That is a perfectly respectable way to look at linear

algebra, but it is not the only way. In this chap-

ter, I want to focus on a higher level of abstrac-

tion. (Learning to handle abstractions is one of the

main goals of this course, in fact!) The purpose

of abstraction is to allow us to understand what is

REALLY going on with some of the things we did,

and also to show that there are many DIFFERENT

ways of thinking about mathematical tools. There

is no single RIGHT WAY to think about linear al-

gebra, despite what some people might try to tell

you. So brace yourself: this chapter (and the next

one) are (intentionally) VERY different from the

first one.

1. WHAT IS A VECTOR?

The basic idea is that vectors (the kind with which

we are familiar) are things that can be multiplied

by numbers and added together. So it makes sense

to generalise this: at a more abstract level, a VEC-

TOR SPACE is any set of things that can be mul-

tiplied by numbers and added together.

EXAMPLE 2.1: The set of all real (or complex)

numbers obviously consists of things that can be

multiplied by numbers and added together. So

numbers of either kind are vectors.

EXAMPLE 2.2: Think of the set of all cubic poly-

nomials with real number coefficients. Of course

it makes sense to multiply them by real numbers:

7 × (x3 + 3x + 9) = 7x3 + 21x + 63

and to add them up:

(x3 + x2) + (2x3 + 3x2) = 3x3 + 4x2,

so they are just like vectors. That being so, we just

call them vectors even though they don’t seem to

have a direction or any obvious length.

EXAMPLE 2.3: Think of the set of all 2 × 2 ma-

trices. They are also vectors in this sense! You

can multiply them by numbers and you can add

and subtract them. That’s right, matrices are vec-

tors....

EXAMPLE 2.4: The set of all 1×n matrices, some-

times called the row vectors, is a very important

vector space in this course, almost as important

as the more familiar n × 1 matrices, the column

vectors.

EXAMPLE 2.5: Let f (x, y, z) be any smooth func-

tion of the Cartesian coordinates x, y, z in three di-

mensional space, let u be any three-dimensional

real vector, and let ∇u be the mapping from the

set of such functions to itself defined by taking the

directional derivative in the direction* of u. (If you

haven’t seen this in calculus yet, just trust me: it

is a gadget that measures the rate of change of

* Some people insist that u has to be a unit vector. I am not
one of those people.
a function in the direction of u. Or, better than

trusting me, trust wikipedia.) It can be shown that

( )
∂f ∂f ∂f
∇u f = u · i + j + k ,
∂x ∂y ∂z
where u is any real three-dimensional vector, f is

any differentiable function of x, y, z, and · is the

dot product.

Then the set of all such mappings can be turned

into a real vector space. (Hint: what is ∇u + v f ?)

Notice that ∇u is a very complicated thing, a map-

ping from a rather abstract set to itself, defined

using epsilons and deltas and all that analysis jazz.

And yet we find that these complicated things be-

have like vectors! Since that is so, maybe it might

be a good idea to say that they are vectors. Be-

cause then we can do with them all of the things

we did in Chapter 1. This turns out to be a ridicu-

lously powerful idea.

Let’s try to make sense of this idea in an exact

way.

2. FORMAL DEFINITION

In this course, the word “scalar” just means a num-

ber which can be either real or complex. We write

F to mean either the set R of all real numbers, or

the set C of all complex numbers. So scalars are

just elements of F .

If n is a positive integer, then a LIST of length

n is just an ordered set of n objects. A scalar

list is a list of scalars.

  For scalar lists we use a
1
 
special notation:  0  is a scalar list of length
3
3. Lists differ from sets in that sets don’t care

about ordering. Similarly, you can have an ordered

set which consists of lists.... that is, a list of lists!

And so on. (But for non-scalars we won’t usually

use this special notation.)

DEFINITION 2.6: F n is the set of all lists of length

n consisting of elements of F .

In other words, these things are just what we nor-

mally call “column vectors”. What we are learning

here is that [a] this is how to define these things

in a precise way, and [b] the idea works for any n,

not just 2 and 3.

Multiplication of a scalar into a list is defined in

the obvious way. Adding and subtracting lists is

also defined in the obvious way. The zero list is

defined in the obvious way for each n. It has the

expected properties. Of course, the number zero is

very different from the zero list. But the expected

things happen: the number zero multiplied into any

list gives the zero list. And so on.

VECTOR SPACES are just sets that are given

properties to make them resemble F n. That is,

you should be able to multiply them by scalars and

add and subtract them.

DEFINITION 2.7: Addition for a set V is a defi-

nite mapping that takes pairs of elements of V to

another element of V . (Still more formally: it is a

mapping from the CARTESIAN PRODUCT V × V

to V .) Notation: if u and v are in V , then the pair

(u, v) is mapped to u + v.

ADVICE 2.8: Don’t think of u and v as numbers

or column vectors or matrices: they could be any-

thing. Try to cultivate the habit of thinking ab-

stractly. The plus sign is just a notation, nothing

more.

DEFINITION„ 2.9: Scalar multiplication for a set

„ Now we will start to use standard pure-mathematics nota-
V is a definite mapping that takes pairs of the form

(a, u), where a is a scalar and u ∈ V , to an element

of V . (Formally: it is a mapping F × V → V .)

Notation: (a, u) is mapped to au.

DEFINITION 2.10: A VECTOR SPACE is a set V

with an addition and a scalar multiplication, such

that

[1] The addition is commutative: u + v = v + u

∀u, v ∈ V

[2] The addition and scalar multiplication are asso-

ciative: (u + v) + w = u + (v + w) ∀u, v, w ∈ V , and

(ab)v = a(bv)∀a, b ∈ F , ∀v ∈ V.

[3] There is an additive identity: ∃0 ∈ V such that

v + 0 = v ∀v ∈ V . This is called the zero vector.

tion for things: ∀ means “for all”, ∈ means “is an element
of”, ∃ means “there exists”. Don’t worry, you will get used
to it.
[4] Every v ∈ V has an additive inverse: ∀v ∈ V, ∃w ∈

V such that v + w = 0.

[5] Multiplicative identity: every v ∈ V satisfies

1v = v, where 1 is the number 1 ∈ F .

[6] Multiplication is distributive both ways: that is,

a(u + v) = au + av and (a + b)u = au + bu ∀a, b ∈ F

and ∀u, v ∈ V .

Elements of vector spaces are called vectors; as we

have stressed, it doesn’t matter whether they look

like “vectors” in the sense of Chapter 1.

If F is the real numbers, the vector space is said to

be “real”; if F is the complex numbers, then V is

said to be a complex vector space. This is an idiotic

terminology (it’s the coefficients that are real or

complex, not the vectors) but that’s mathematical

life.
Of course, with the obvious definitions of adding

elements and multiplication by scalars, F n can be

turned into a vector space, and from now on “F n”

MEANS this vector space.

The above axioms are designed to capture anything

that sufficiently resembles F n. Because of this, all

of the things we know about F n are equally true

of general vector spaces. For example, you can

prove (see Tutorials) that a vector space cannot

have two different zero vectors, and that a vector

cannot have two different additive inverses. (So we

can give the unique additive inverse of u a name,

−u. And then show that (−1)u = −u. And then

define u − v = u + −v. And so on.)

3. SUBSPACES

DEFINITION 2.11: A subset U of a vector space

V is called a SUBSPACE if U is a vector space,

with the same scalar multiplication and addition as

in V .

Most subsets of a vector space aren’t subspaces.

For example, the set consisting of any single vec-

tor (other than the zero vector) is obviously not a

vector subspace. But of course some are: for ex-

ample, the subset of F 2 consisting of all vectors of

( )
a
the form is easily seen to be a subspace. The
0
point is that multiplication by a scalar and addition

of vectors in this subset do not take you outside

that subset.

DEFINITION 2.12: Suppose U1 and U2 are sub-

SETs of a vector space V . Then the SUM of these

subsets, denoted U1 + U2, is defined as the subset

of V consisting of all vectors of the form u1 + u2,

where u1 ∈ U1, and u2 ∈ U2.

It’s easy to see that the sum of two subspaces is

again a subspace. In fact these things behave in a

particularly nice way in the following special case.

DEFINITION 2.13: Let U1 and U2 be subspaces

of a vector space V . Suppose that the intersection

U1 ∩ U2 = {0} (Note: this isn’t the empty set!).

Then U1 + U2 is called the DIRECT SUM of U1

and U2. Notation: the direct sum of U1 and U2 is

denoted U1 ⊕ U2.

The nice thing about direct sums is the following.

Suppose that you have expressed a vector v ∈ V as

u1 + u2 ∈ U1 ⊕ U2. Then that expression is unique,

that is, there is no other way of doing this. To see

this, suppose that v = w1 + w2, where w1 and w2

are some other vectors in U1, U2 respectively. Then

Strictly, it is called the internal direct sum. There is also
an external direct sum. The difference is not important (for
us).
clearly u1 − w1 = w2 − u2, so u1 − w1 ∈ U1 (because

it’s a subspace) and u1 −w1 ∈ U2 (because it’s equal

to something in U2!) so it is in the intersection, and

so it must be the zero vector; therefore u1 = w1

and similarly u2 = w2. So the expression is unique.

EXAMPLE 2.14: Let U1 be the subspace of F 2

( )
a
consisting of vectors of the form , and let
0
U2 be the subspace consisting of vectors of the
( )
0
form . Then the intersection obviously con-
b
sists of the zero vector alone, and in fact F 2 =

U1 ⊕ U2. And indeed it is clearly true that vectors

can be expressed as sums of this form in only one

way...otherwise our lives in Chapter 1 would have

been much more interesting.

EXAMPLE 2.15: Consider F n, and define the dot

product of two vectors in just the same way as in

two and three dimensions; that is, by multiplying

the components together and then adding them

up. Let U be any subspace of F n; since it has to

contain the zero vector, you can think of it as a

(hyper)plane passing through the origin in the usual

picture of F n. Let U ⊥ be the set of all vectors in

F n which are perpendicular to every vector in U .

Then F n = U ⊕ U ⊥.

To see this, let v ∈ F n, and think of it as the

position vector of a point P . Let Q be the point

in U that minimizes the function (of position in U )

which gives the distance from P to all points in

U . Then let u be the position vector of Q; clearly

u ∈ U, and v − u is perpendicular to every vector

in U . (The existence of Q, and the fact that v − u

is perpendicular to every vector in U , are exercises

in calculus, so I won’t discuss them here§.) Let

§ See the wikipedia page about “Hilbert projection theorem”,
finite dimensional case.
w = v − u. Then of course v = u + w, where u ∈ U

and w ∈ U ⊥. That is, V = U + U ⊥. Since the zero

vector is the only vector perpendicular to itself, U

and U ⊥ only intersect in the vector space consisting

of nothing but the zero vector. Thus if we take the

sum, it is indeed a direct sum.

Notice that we are really using the dot product

here: without it, we would not know how to de-

fine “the distance from P to U ”, and we would

not know what “perpendicular” means! So this

whole argument wouldn’t make sense in a general

abstract vector space, where you don’t in general

have anything analogous to the dot product (but

see Chapter 5!)

4. VECTOR SPACE ISOMORPHISMS

One often finds that two vector spaces are “some-

how the same” even though, strictly speaking (as

sets) they aren’t really the same thing. For ex-

ample, think about the vector space of all 2 × 2

matrices. You can write them like this:

( ) ( ) ( ) ( )
1 0 0 1 0 0 0 0
a +b +c +d .
0 0 0 0 1 0 0 1
It seems intuitively clear that, somehow, these ma-

trices behave as if they were in F 4; because, as far

as adding and multiplying by scalars are concerned

(and we know that these things are the essence of

what it means to be a vector space) they seem to

be the “same”, or to “contain the same amount

of information,” even though of course they really

aren’t the same. Let’s see how to make this more

definite.

Let S and T be two sets. Recall that a mapping

F : S → T is called a SURJECTION if, for any

t ∈ T , there exists s ∈ S such that F (s) = t. (This

means that EVERYTHING in T can be expressed

in the form F (s).)

It is called an INJECTION if, for all s1, s2 ∈ S,

F (s1) = F (s2) implies that s1 = s2. (This means

that, if it is possible to express a vector in T in the

form F (s), THERE IS ONLY ONE WAY TO DO

THIS: that is, s is UNIQUE.)

It is called a BIJECTION if it is both surjective

and injective. Remember that F has an INVERSE

if (and only if) it is a bijection.

Now let U and V be vector spaces. A mapping

Φ : U → V is said to be a VECTOR SPACE HO-

MOMORPHISM if it is as nice as possible: ∀ u, v ∈

U, ∀ a ∈ F ,

Φ(u + v) = Φ(u) + Φ(v)

Φ(au) = aΦ(u).
DEFINITION 2.16: If there exists a vector space

homomorphism from U to V which is a bijection,

then we call the homomorphism an ISOMORPHISM,

and we say that the two vector spaces are ISO-

MORPHIC.

This is what we mean by the idea that two different

vector spaces are “so similar that they might be

twins”: we mean that they are isomorphic. For

example, I leave it to you to show in detail that

the mapping
 
( ) a
 
a b  b 
→ 
c d  c 
d
is indeed an isomorphism. They are NOT the

same.... but they are so similar that they behave

in much the same way, as vector spaces.

Unfortunately, everyone tends to be a bit sloppy

about this, and sometimes people say that two

things are the same when they really mean that

they are isomorphic.

EXAMPLE 2.17: In EXAMPLE 2.14, U1 is clearly

isomorphic to F 1, and actually so is U2! So peo-

ple usually say that “F 1 ⊕ F 1 is isomorphic to F 2”.

Strictly speaking, this does not make sense, be-

cause F 1 is NOT really a subspace of F 2... it isn’t

even a subSET! But we still talk like this.

For example, F n+m has two subspaces isomorphic

to F n and F m respectively, so everyone writes

F n ⊕ F m ↔ F n+m,

where ↔ means, “is isomorphic to”, even though

this isn’t really right. Beware. (The strictly correct

statement is that F n+m contains two subspaces,

one of which is isomorphic to F n, and one which is

isomorphic to F m, and F n+m is isomorphic to the

direct sum of those two subspaces.)

Now as you might expect, any vector space is iso-

morphic to itself. In fact, with one exception they

are isomorphic to themselves in INFINITELY MANY

DIFFERENT WAYS!

SIMPLE BUT VERY IMPORTANT FACT 2.18:

Apart from the boring vector space with just one el-

ement, which is isomorphic to itself in just one way,

vector space isomorphisms ARE NOT UNIQUE:

that is, if there is one isomorphism from one vec-

tor space to another, then there are (infinitely!)

many others. Trivial example: let c be any non-

zero number, let V be a vector space, and consider

the mapping v → cv for any v ∈ V . This is obvi-

ously an isomorphism from V to itself, for infinitely

many different c.

But if vector spaces are isomorphic to themselves

in many ways, then they are isomorphic to other

vector spaces in many ways. To see this, let Φ

be an isomorphism of V to itself, and let Ψ be an

isomorphism from V to another vector space W .

Then for each Φ, the composite map Ψ ◦ Φ is an

isomorphism from V to W . (Draw a picture!) Of

course if there are infinitely many Φ, then there are

infinitely many Ψ ◦ Φ!

DEFINITION 2.19: A vector space is said to be

FINITE-DIMENSIONAL OVER F if it is isomor-

phic to F n for some (finite!) integer n.

No finite-dimensional vector space can be isomor-

phic to F n for two different values of n (Tuto-

rial question), so this number is a fixed property

of the vector space. It is called the DIMENSION

of the vector space. In our example above, we

see that the vector space of all 2 × 2 matrices is

four-dimensional. (Notice however that things can

depend on whether F is the real numbers or the

complex numbers; for example, the vector space of

complex numbers is one-dimensional over the com-

plex numbers, but of course two-dimensional over

the real numbers. So we should really say things

like “two-dimensional over the reals”, or “two real-

dimensional” etc. In practice this doesn’t cause

any trouble. Usually.)

Of course, not every vector space is finite-dimensional:

for example, the set of all polynomials with real co-

efficients is a vector space, but it’s pretty clear that

there no isomorphism with F n for any n (the de-

gree of the polynomial can be chosen to be greater

than any given integer). We say (obviously) that

it is INFINITE-DIMENSIONAL.

In this course, we deal almost exclusively with finite-

dimensional vector spaces. In fact, if I ever for-

get to say, “finite-dimensional”, you should assume

that that is what I mean. Infinite-dimensional vec-

tor spaces can behave very differently from their

finite-dimensional brothers, but many properties of

finite-dimensional vector spaces do carry over to

the infinite-dimensional case. Many of the things

we will discover in this course do so (especially in

Chapter 3) so those of you who need the infinite-

dimensional case (such as quantum-mechanical peo-

ple) will still find it useful. If you are careful.

5. SPANNING, LINEAR INDEPENDENCE,

BASIS: OLD-FASHIONED VERSION

Remember, a list is just an ordered set...of any-

thing. So, in a given vector space V , we can con-

sider a list of vectors. This will look like {v1, v2, ...}.

We will be a bit sloppy and denote lists of vectors

like this: vi, where i = 1, 2, 3, etc.

DEFINITION 2.20: A LINEAR COMBINATION of

the elements of a list of vectors vi is just any vector

(which may or may not lie in the list!) of the form

a1v1 + a2v2 + a3v3..., where the ai, with i = 1, 2, 3

etc are scalars. (Of course, you won’t confuse the

notation ai with powers, correct?) We will write

∑ i
this as a vi. Sometimes we might even get really

lazy and just write this as aivi, with the under-

standing that a sum has to be taken. The symbol

∑
is pretty, but it doesn’t really add anything to

our understanding....from now on, when you see

REPEATED indices as in this case, you should as-

sume that a sum is intended. FROM NOW ON,

∑ i
aivi MEANS a vi. Yes, I know this notation looks

like a disaster waiting to happen, but surprisingly

it works extremely well. Trust me.

DEFINITION 2.21: The SPAN of a list of vectors

is defined to be the set of all linear combinations

of the vectors in the list.

EXAMPLE 2.22: If the list consists of one vector,

then its span is just the set of all multiples of that

vector.
{( ) ( )}
1 0
EXAMPLE 2.23: The span of the list ,
0 1
is all of F 2.

FACT 2.24: Given any list of vectors in V , the span

of that list is a subspace of V (but note that V is

of course a subspace of itself; if the span of a list

equals V , we say that the list spans V ).

DEFINITION 2.25: A list of vectors vi in V is said

to be LINEARLY INDEPENDENT if the equation

∑ i
a vi = 0 implies that ai = 0 for every i.

This is nice, because it means that the elements

of the span of a linearly independent vi can be ex-

pressed as a linear combination in a UNIQUE way:
∑ i
if a certain vector in the span, a vi can ALSO be
∑ i ∑ i
expressed as b vi, then of course (a − bi)vi = 0,

and by the definition of linear independence this

means ai = bi for every i. (Note: if I had dropped

∑
everywhere in that last sentence, it would still

be understandable, right?)
{( ) ( )}
1 0
EXAMPLE 2.26: The list , is linearly
0 1
( ) ( ) ( )
1 0 0
independent: obviously a +b =
0 1 0
implies a = b = 0.

DEFINITION 2.27: A BASIS of a vector space V

is any list of vectors in V which BOTH spans V

AND is linearly independent.

In other words, every vector is covered by this pro-

cess, and it happens in a unique way. Ring any

bells? Doesn’t this remind you of the definition of

a vector isomorphism? Yes, it does! We will come

back to this.

FACT 2.28: Every finite-dimensional vector space

has a basis.

We will see the real meaning of this statement at

the end of this chapter. In fact, when you under-

stand what a basis really is, you will see that this

FACT is really quite trivial¶.

{( ) ( )}
1 0
EXAMPLE 2.29: The list of vectors ,
0 1
spans and is linearly independent in F 2, so it is a

basis. As you know.

COMMENT 2.30: (VERY IMPORTANT): Apart

from the boring vector space that consists of just

¶ Infinite-dimensional vector spaces also always have a basis,

but in this case this fact is very NON-trivial.
one element (the zero vector, of course!), NO VEC-

TOR SPACE EVER HAS JUST ONE BASIS! In

fact, all vector spaces other than the boring one

have INFINITELY MANY different bases!

To repeat: bases are useful because every vector

in the vector space can be expressed as a linear

combination of the elements of the basis, AND

this can be done in only one way for a given basis.

BUT it is crucial to remember that many bases are

available. You can choose them to suit your own

convenience. As you know.

6. TIME TO TAKE THAT RED PILL:

WHAT BASES REALLY ARE

According to the above discussion, a basis is just

a list of vectors in some vector space V , so the

strict notation for it is {z1, z2, z3....}, that is, we

should use set notation, because a list is a kind

of set. But usually we don’t do that, we just write

z1, z2, z3.... with the understanding that it’s a list of

vectors. As mentioned earlier, actually I will often

be even lazier than this and talk about the basis

“zi”, where the understanding is that the subscript

could be any integer between 1 and some finite

value. Sometimes I will be EVEN LAZIER and

just talk about “the basis z”, meaning the list zi.

I won’t get any lazier than that. This works bet-

ter than you might expect; the meaning is nearly

always clear from the context.

The most important things about bases is the fact

that, given a fixed basis z or zi, we can [a] express

any vector v in a finite-dimensional vector space

V as v = aizi, where of course the sum is finite,

and [b] this expression is unique. The numbers

ai are called the COMPONENTS of v relative to

this basis. (The uniqueness is what allows us to

talk about THE components.) This is indeed ex-

tremely important, because it gives us a language

in which to talk about vectors: without bases, we

could not really do this! (My convention is that

names of components come from the beginning of

the alphabet, while names of bases come from the

end.)

Now comes the abstraction: instead of thinking

about a basis z as a collection of vectors, think

about it as a MAPPING: it is a gadget that takes

a (finite) list of numbers and turns that list into a

vector in a given vector space. We can of course

arrange the list of numbers into an element of F n,

for some n. So, a little more formally:

A basis of a finite-dimensional vector space V de-

fines a MAPPING z that turns elements of F n into

vectors in V . So I can write things like z(a), where

a ∈ F n, and this just means “the vector in V that

has components ai relative to the basis zi”. Well,

as usual in Linear Algebra, we are lazy about brack-

ets (as we are generally) and we normally write z(a)

as just za. Notice that z(a) = za = aizi.

Clearly the map z is surjective (every vector can

be expressed in terms of components: the basis

SPANS) and injective (this can be done only in one

way: the basis is LINEARLY INDEPENDENT). So,

as a mapping, a basis is a bijection. Furthermore,

you can easily see that, if a1, a2 ∈ F n and c ∈ F ,

then z(a1 + a2) = za1 + za2, and z(ca1) = cza1.

So in fact a basis is a vector space isomorphism!

This is very nice. The usual familiar statement, “A

basis spans,” just means that the mapping is sur-

jective. And the familiar statement, “A basis is lin-

early independent” just means that the mapping is

injective! So thinking about a basis as a mapping is

extremely natural. To put it more strongly: SPAN-

NING AND LINEAR INDEPENDENCE ARE JUST

WORDS THAT MEAN “SURJECTIVE” and “IN-

JECTIVE”!

On the other hand, vector space isomorphisms de-

fine bases in the old-fashioned† sense, a fact that

can be seen with the help of the following defini-

tion.

DEFINITION 2.31: We define {ei} to be the CANON-

ICAL BASIS of F n, that is, e1 has 1 in the first slot,

† Despite being old-fashioned, the traditional way of thinking
about bases is still very useful, as you will see in the next
Chapter. The point is that there is another way, which is
also very useful. “Old” ̸= “Useless”. Think of the music of
Beethoven. Other examples have surely occurred to you...
then zeroes everywhere else, e2 has 1 in the second

slot and zeroes everywhere else, and so on (so the

familiar i, j, k vectors are the canonical basis of

F 3). (CANONICAL just means, “very special”, or

“really cool.”)

Now suppose that z : F n → V is a basis (in the

“new” sense, that is, it is a vector space isomor-

phism). Then the list of vectors zi ≡ z(ei) is a basis

for V (in the old-fashioned sense, that is, it is a list

of vectors). (Please convince yourself of this, it’s

quite easy.) So there is a one-to-one correspon-

dence between bases (defined in the old-fashioned

way) and vector space isomorphisms of the form

z : F n → V . Anything you can do with one, you

can do with the other.

So what is a basis, really? It’s just a specific ex-

ample of the vector space isomorphism between

F n and V ... the existence of which is the defini-

tion of being finite-dimensional! So from this point

of view the existence of a basis is just a matter of

definition (in the finite-dimensional case).

This way of thinking about bases is VERY help-

ful, because it means that, by picking a basis, we

can turn any problem about abstract vectors into a

much more concrete problem about F n (since the

basis allows us to go back and forth between F n

and any finite-dimensional vector space).

For example:

FACT 2.32: The number of vectors in an old-

fashioned basis of a finite-dimensional vector space

is always equal to the dimension.

This is true because there are exactly n different

vectors ei in F n, so (since z is an isomorphism)

there are n different vectors z(ei) = zi.

You can see how nice this is! A lot of proofs which

are complicated in the textbooks are actually very

simple when you think this way. Another example:

FACT 2.33: Suppose U1 and U2 are subspaces of a

finite-dimensional vector space V , with intersection

consisting only of the set consisting of the zero

vector. Then dim(U1 ⊕ U2) = dim(U1) + dim(U2).

You can see why this is so by letting n = dim(U1), m =

dim(U2), so that U1 is isomorphic to F n, and U2

is isomorphic to F m. Now as we discussed earlier,

F n+m has two subspaces, let’s call them A and B,

such that A is isomorphic to F n, and B is isomor-

phic to F m: they look like this:

( ) ( )
a 0
A= , B= ,
0 b
where a consists of n numbers and b consists of

m numbers. We know that F n+m is isomorphic to

the direct sum of A and B. Now remember that

since U1 ⊕ U2 is a direct sum, any element in it can

be expressed unambiguously in the form u1 + u2,

we have a well-defined (that is, unambiguous) iso-

morphism from U1 ⊕ U2 to F n+m given by mapping

( )
a
u1 +u2 to , where a is the image of u1 and b is
b
the image of u2 with respect to the two given iso-

morphisms. Hence dim(U1 ⊕ U2) = dim(F n+m) =

m + n = dim(U1) + dim(U2).

OK, there’s a bit of notation there, but it’s pretty

clear what is going on here: the statement is rather

obvious in F n+m, so it must also be true in V !

Here is another example of the same philosophy.

FACT 2.34: Suppose U is a subspace of a finite-

dimensional vector space V . Then V has another

subspace, W , such that V = U ⊕ W .

REMARK 2.35: Informally, we are sort of saying

that W is what you get when you “subtract” U

from V (!) HOWEVER, be careful: W isn’t unique,

there are many choices. (Think of R2, and let U

be the vectors on the x axis. Let w be any vector

not parallel to the x axis, and let W be the vector

subspace consisting of all multiples of w. Then

R2 = U ⊕ W.)

To see why FACT 2.34 is true, let z be any basis of

V , so z is an isomorphism from F n to V for some n.

Now consider z −1(U ); because z is an isomorphism,

it is a subspace of F n, so by EXAMPLE 2.15 there

exists a subspace X of F n, namely (in the notation

( )⊥
−1
we used there) X = z (U ) , such that

F n = z −1(U ) ⊕ X.
You can show that a vector space isomorphism

maps direct sums to direct sums (meaning ϕ(U1 ⊕

U2) = ϕ(U1) ⊕ ϕ(U2) for any isomorphism ϕ and

any vector subspaces U1 and U2 in a larger vector

space: see Tutorial 3). So, since V = z(F n),

V = U ⊕ z(X),

and now, by defining W = z(X), we are done.

Again, the idea here is that this kind of “subtrac-

tion” is possible in F n... but V is isomorphic to

F n, so the same thing must be possible for V !

REMARK 2.36: Notice that the discussion in EX-

AMPLE 2.15 depended heavily on having a dot

product. In an abstract vector space V , we don’t

have anything analogous to a dot product. For

example, take the vector space of all smooth real

functions. What is the angle between ex and sin(x)?

Not too obvious, right? And the same goes even

in the finite-dimensional case: we don’t have a dot

product for a general vector space. But by us-

ing the vector space isomorphism idea, we can still

transfer the conclusion in EXAMPLE 2.15 over to

the general abstract case! The conclusion is still

true in V even though we can’t prove it there in

the same way as in EXAMPLE 2.15. Cool, right?

SUMMARY 2.37: A general, abstract finite-dimensional

vector space V is just a copy of F n for some n. The

copying is done by means of vector isomorphisms

z, which map from F n to V . The isomorphisms

are almost never unique. They are what blue-pilled

people mean when they talk about bases. Anything

in V can be mapped to F n by choosing a specific

z. And that’s all there is to it.