MATH 146 January 9 Section 2

Pseudo-Definition. A field is a “number system” F having:

(1) elements 0,1 (and possibly others). Note: we require that 0 ̸= 1.
(2) operations + and ×, secondary operations − and −1 (the last defined for all nonzero elements)
and satisfying the “obvious” algebraic laws. (See the end of these notes for the official definition.)
Example 1. R, C, Q, Zp (p a prime) are fields. Another example is R(x), the field of all rational
functions p(x)/q(x) where p(x) and q(x) are polynomials over R (and q(x) ̸= 0).
Example 2. N, Z, Zn (n composite) are not fields. The class of all ordinals (under ordinal addition
and multiplication) is not a field.
Let F be a field.
Definition. A vector space over F is a set V on which two operations
• addition, V × V → V , denoted x + y
• scalar multiplication, F × V → V , denoted ax
are defined , and such that the following conditions hold :
For all x, y, z ∈ V and a, b ∈ F:
(VS 1) x+y =y+x
(VS 2) (x + y) + z = x + (y + z)
(VS 3) There exists a “zero vector” in V , denoted 0, which satisfies x + 0 = x for all x ∈ V .
(VS 4) For every x ∈ V there exists u ∈ V satisfying x + u = 0.
(VS 5) 1x = x
(VS 6) (ab)x = a(bx)
(VS 7) a(x + y) = ax + ay
(VS 8) (a + b)x = ax + bx
To define a vector space, you must specify the set and the two operations.
To prove that a set with two operations is a vector space, you need to verify the 8 conditions.
Example 3. Rn is the set of all n-tuples (a1 , a2 , . . . , an ) of real numbers. Addition and scalar
multiplication (by real numbers) on Rn are defined “coordinate-wise,” i.e.,
def
(a1 , a2 , . . . , an ) + (b1 , b2 , . . . , bn ) = (a1 +b1 , a2 +b2 , . . . , an +bn )
def
c(a1 , a2 , . . . , an ) = (ca1 , ca2 , . . . , can ).
Claim. Rn with coordinate-wise addition and scalar multiplication is a vector space over R.
Proof sketch. Must verify that each of (VS 1)–(VS 8) holds. □
Example 4. More generally, for any field F, the set Fn = {(a1 , a2 , . . . , an ) : a1 , . . . , an ∈ F} of
all n-tuples from F, with coordinatewise addition and scalar multiplication (calculated in F), is a
vector space over F.
Example 5. For any field F, the set FN = {(a1 , a2 , . . . , an , . . .) : ai ∈ F} of all infinite sequences
over F, with coordinatewise addition and scalar multiplication (calculated in F), is a vector space
over F.
Example 6. Let F = C. Let V = {o} where o is the apple I brought to class. Defining addition
and scalar multiplication in the only possible way, V is a vector space over C.

Here is our official definition of “field” (from Appendix C of Friedberg, Insel & Spence).
Definition. A field is a set F on which two operations + and · (called addition and multiplica-
tion) are defined (for all possible pairs in F, and always producing an element of F), and having
two privileged elements 0 and 1 with 0 ̸= 1, such that the following laws hold: for all a, b, c ∈ F,
(F 1) a+b=b+a
a·b=b·a
(F 2) a + (b + c) = (a + b) + c
a · (b · c) = (a · b) · c
(F 3) 0+a=a
1 · a = a.
(F 4) For every a ∈ F there exists x ∈ F with a + x = 0.
For every a ∈ F with a ̸= 0 there exists y ∈ F with a · y = 1.
(F 5) a · (b + c) = (a · b) + (a · c)
In any field F, we define −a to be the unique solution x to a + x = 0. Note that (F 4) guarantees
the existence of a solution, and uniqueness can be deduced from (F 1)–(F 3).
In any field F, if a ∈ F and a ̸= 0, then we define a−1 to be the unique solution x to ax = 1.
Note that (F 4) guarantees the existence of a solution, and uniqueness can again be deduced from
(F 1)–(F 3).

2
MATH 146 January 11 Lecture notes

Announcements:

More examples of vector spaces (over a field F):

(1) Fix m, n ≥ 1. Mm×n (F) is the set of all m × n matrices with entries from F. Define addition of
matrices in the usual way (componentwise); also define scalar multiplication componentwise:
a11 a12 · · · a1n ca11 ca12 · · · ca1n
   
 a21 a22 · · · a2n  def  ca21 ca22 · · · ca2n 
c
 ... .. ..  =  .. .. ..  .
. .   . . . 
am1 am2 · · · amn cam1 cam2 · · · camn
This is a vector space. What is the zero vector? Note: though you know how to multiply matrices,
the product operation is not itself part of the vector space. (Also, if m ̸= n then you can’t multiply
two m × n matrices.)
(2) Let F(R, R) be the set of all functions f : R → R. We can turn F into a vector space over R, using
the usual operations of adding two functions, and scaling a function by a real number.
(3) More generally, fix a nonempty set D. Let F(D, F) be the set of all functions f : D → F. We can
turn F(D, F) into a vector space by defining addition and scalar multiplication “pointwise.” I.e., if
f, g ∈ F(D, F), then we need to define f + g and this sum must be a function from D to F. We do
this by setting, for each x ∈ D,
def
(f + g)(x) = f (x) + g(x).
Scalar multiplication is defined similarly: if f ∈ F(D, F) and c ∈ F, then the function cf is defined
by setting, for each x ∈ D,
def
(cf )(x) = c · f (x).
With these two operations, F(D, F) is a vector space over F.
(4) Fix n ≥ 0. Pn (F) denotes the set of all “formal polynomials” in the variable x, of degree at most
n, using coefficients from F. Thus
Pn (F) = {a0 + a1 x + a2 x2 · · · + an xn : a0 , a1 , . . . , an ∈ F}.
“Formal” means that two polynomials are considered to be equal if and only if they have exactly
the same coefficients (not when they define the same polynomial functions F → F).
Addition of polynomials in Pn (F) is defined “term-wise” in the usual way. Similarly, multiplica-
tion of a polynomial by a scalar c ∈ F is defined by multiplying the coefficient of each term by c.
Pn (F) with these operations is a vector space over F.
Note that
F = P0 (F) ⊂ P1 (F) ⊂ P2 (F) ⊂ · · · ⊂ Pn (F) ⊂ Pn+1 (F) ⊂ · · ·
We let P(F) be the set of all formal polynomials (of all degrees) with coefficients from F. Thus
∞
[
P(F) = Pn (F).
n=0

P(F) with addition and scalar multiplication defined term-wise is a vector space over F.
Remark. A more standard notation for P(F) is F[x], but we will use P(F) in this course.
Next: some basic facts true of all vector spaces.
Theorem 1.1 (Cancellation Law). Let V be a vector space. If x, y, u ∈ V and x + u = y + u, then x = y.
Proof. Assume x + u = y + u. Let 0 be the zero vector of V (given by (VS 3). By (VS 4), there exists
z ∈ V with u + z = 0. Then
x = x+0 (VS 3)
= x + (u + z) by choice of z above
= (x + u) + z (VS 2)
= (y + u) + z assumption
= y + (u + z) (VS 2)
= y+0 choice of z above
= y (VS 3).
□
Corollary 1. Suppose V is a vector space. There is only one vector in V that can be the zero vector.
Proof. Suppose 01 , 02 ∈ V both satisfy (VS 3). Thus x + 01 = x + 02 = x for all x ∈ V . Use (VS 1) to flip
this to get 01 + x = 02 + x. Then 01 = 02 by the Cancellation Law. □
Corollary 2. Suppose V is a vector space and x ∈ V . There is only one vector u ∈ V satisfying x + u = 0.
Proof. Like the proof of Corollary 1; exercise. □
Definition. Let V be a vector space and x, y ∈ V .
(1) −x denotes the unique vector u ∈ V satisfying x + u = 0.
(2) x − y denotes x + (−y).
Theorem 1.2. Suppose V is a vector space over F, x ∈ V , and a ∈ F.
(1) 0x = 0.
(2) (−a)x = −(ax) = a(−x).
(3) a0 = 0.
Remark. Parts (1) and (2) are proved in the text. Parts (1) and (3) are also proved in Prof. Tatarko’s
notes.
Also note the overloaded notation in the statement. For example, in part (1), the first 0 is the scalar
0 ∈ F, while the second 0 is the zero vector in V . In part (2), the first − means minus in F, while the
second and third mean the minus in V defined above.
Here is a proof of (1), where I will use bold font for vectors, nonbold for scalars: In the field F we have
0 + 0 = 0; thus 0x = (0 + 0)x = 0x + 0x by (VS 8). On the other hand, we have 0x = 0x + 0 = 0 + 0x by
(VS 3) and (VS 1). Putting these together gives 0x + 0x = 0 + 0x. Now apply the Cancellation Law to
get 0x = 0.

2
MATH 146 January 13 Section 2

Assume V is a vector space over F and x1 , . . . , xn ∈ V . Because of (VS 2), we can (and do) write
expressions like x1 + x2 + · · · + xn without declaring where the brackets go. And we could (though we
won’t) prove true facts like
x1 + x2 + · · · + xn = xσ(1) + xσ(2) + · · · + xσ(n) for any permutation σ of {1, 2, . . . , n}
c(x1 + x2 + · · · + xn ) = cx1 + cx2 + · · · + cxn for any scalar c ∈ F
−(x1 + x2 + · · · + xn ) = −x1 − x2 − · · · − xn
Going forward, you can freely use these facts, except when told not to.

Definition. Let V be a vector space over a field F, and let W ⊆ V .

(1) W is closed under addition (of V ) if x, y ∈ W =⇒ x + y ∈ W .
(2) W is closed under scalar multiplication (of V ) if x ∈ W and a ∈ F =⇒ ax ∈ W .
Definition (Different from our text!). Let V be a vector space over F. A subset W ⊆ V is a called
subspace of V if
(1) W is closed under the operations of V , and
(2) W ̸= ∅.
Lemma. Suppose V is a vector space over F and W is a subspace of V . Then W , with the operations of V
restricted to W , is itself a vector space over F.
Proof. First, the restrictions to W of the operations of V are operations of the right kind (W × W → W
and F × W → W ), because W is closed under the operations of V .
Next, (VS 1), (VS 2), and (VS 5)–(VS 8) follow automatically from their truth in V because of their
logical nature (universally quantified statements) and the fact that the operations of W are just the
restrictions to W of the operations of V . Proving (VS 3) and (VS 4) requires more work. Let +V be the
addition operation of V and let + denote its restriction to W .
(VS 3): since W ̸= ∅, we can choose x ∈ W . Then 0x ∈ W (as W is closed under scalar multiplication).
But 0x is the zero vector 0V of V (by Theorem 1.2(1) applied to V ), so 0V ∈ W . We’ll now show that 0V
is the zero vector of W : for any x ∈ W ,
x + 0V = x +V 0V (+ is the restriction of +V )
=x since (VS 3) is true in V and x ∈ V .

(VS 4): given x ∈ W , recall that −1 ∈ F so (−1)x ∈ W . Applying Theorem 1.2(2) and (VS 5) to V , we
get (−1)x = −(1x) = −x (the additive inverse of x in V ), which proves −x ∈ W . We now show that −x
is also an additive inverse of x in W :
x + (−x) = x +V (−x) = 0V by definition of −x. □
Incidentally, the proof shows that if W is a subspace of V , then:
• W always contains the zero vector of V (which is also the zero vector of W );
• W is closed under additive inverses (calculated in V ). Moreover, if x ∈ W then the additive inverse
of x calculated in V is the same as the additive inverse of x calculated in the vector space W .
Example.
(1) V = R3 as a vector space over R (usual operations). Let W be the following plane in R3 :
W = {(x, y, z) ∈ R3 : x + y + z = 0}.
It’s easy to check that W is a subspace of R3 : i.e., it is a subset of R3 , it is closed under addition
and scalar multiplication, and it is nonempty.
(2) Same V . Let W1 = {(x, y, z) ∈ R3 : x + y + z = 1}. W1 is not a subspace of R3 , for many reasons:
(i) it doesn’t contain the zero element of R3 , so it can’t be a subspace by the observation following
the proof of the Lemma; (ii) W isn’t closed under scalar multiplication; and (iii) W isn’t closed
under vector addition.
(3) Let W2 be the paraboloid
W2 = {(x, y, z) ∈ R3 : z = x2 + y 2 }.
W2 is not a subspace of R3 , as it is not closed under scalar multiplication. (It’s also not closed
under vector addition.)
(4) Given any vector space V :
• V is a subspace of itself.
• {0} is a subspace.
(5) The subspaces of R3 are R3 , all planes through 0, all lines through 0, and {0}.
(6) Let V be C considered as a vector space over R. (So scalar multiplication is R × C → C.)
(a) R is a subspace of V . (Proof: R ̸= ∅, R is closed under (complex) addition, and R is closed
under scaling by real numbers.)
(b) Can you find another subspace of V (other than C and {0})?
(7) Is R2 a subspace of R3 ?
(8) For each field F and n ≥ 0, Pn (F) is a subspace of Pn+1 (F) and is also a subspace of P(F).

2
MATH 146 January 16 Section 2

Theorem 1.4. Let V be a vector space over F. Let W1 , W1 be subspaces of V . Then W1 ∩ W2 is also a
subspace of V .
Proof. We check the definition of being a subspace.
(1) Let x, y ∈ W1 ∩ W2 . Then x, y ∈ W1 so x + y ∈ W1 . Similarly, x, y ∈ W2 so x + y ∈ W2 . Thus
x + y ∈ W1 ∩ W2 , proving W1 ∩ W2 is closed under addition.
(2) Let x ∈ W1 ∩ W2 and c ∈ F. A similar proof shows cx ∈ W1 ∩ W2 .
(3) We have 0 ∈ W1 and 0 ∈ W2 by the earlier Lemma, so 0 ∈ W1 ∩ W2 , proving W1 ∩ W2 ̸= ∅. □
Note: in general, it is not true that the union of two subspaces is a subspace. See Assignment 1.

§1.4

Definition. Suppose V is a vector space over F and x ∈ V .

(1) For v1 , . . . , vn ∈ V , we say that x is a linear combination of v1 , . . . , vn (or of {v1 , . . . , vn }) if
there exist c1 , . . . , cn ∈ F such that x = c1 v1 + · · · + cn vn .
(2) If S ⊆ V is infinite, we say that x is a linear combination of S if x is a linear combination of
some v1 , . . . , vn ∈ S.
Definition 1.5. Suppose V is a vector space over F and ∅ ̸= S ⊆ V . The span of S is the set
def
span(S) = {all linear combinations of S} ⊆ V.
def
Also span(∅) = {0}.
Example. In R3 , the span of {(1, 0, 0), (0, 1, 0)} is
{a(1, 0, 0) + b(0, 1, 0) : a, b ∈ R} = {(a, b, 0) : a, b ∈ R},
the x-y plane in R3 .
Example. In P(R), what is the span of the infinite set S = {x, x2 , x3 , x4 , . . .}? It includes all linear
combinations of finitely many of x, x2 , x3 , x4 , . . .. Thus we get all polynomials of the form
a1 x + a2 x 2 + · · · + an x n , a1 , a2 , . . . , an ∈ R.
In other words, span(S) = {f (x) ∈ P(R) : f (0) = 0}.
span(S) will turn out to always be a subspace. Before proving this, we need the following.
Technical Observation. Suppose V is a vector space, S ⊆ V with S infinite, and x, y ∈ span(S). Then
each of x, y is a linear combination of some finite set of vectors in S, but perhaps not the same finite set.
Not a problem: if
x = a1 u1 + · · · + am um with u1 , . . . , um ∈ S
y = b1 v1 + · · · + bn vn with v1 , . . . , vn ∈ S
then, letting k = |{u1 , . . . , um } ∩ {v1 , . . . , vn }|, without loss of generality we can assume ui = vi for
i = 1, . . . , k and {uk+1 , . . . , um } ∩ {vk+1 , . . . , vn } = ∅ and write
x = a1 u1 + · · · + ak uk + ak+1 uk+1 + · · · + am um + 0·vk+1 + · · · + 0·vn
y = b1 u1 + · · · + bk uk + 0 · uk+1 + · · · + 0 · um + bk+1 vk+1 + · · · + bn vn
and both x, y are now written as linear combinations of the same finite subset {u1 , . . . , um } ∪ {vk+1 , . . . , vn }
of S.
Theorem 1.6. Let V be vector space over F. Let S ⊆ V . Then:
(1) span(S) is a subspace of V .
(2) span(S) contains S, i.e., S ⊆ span(S).
(3) Moreover, span(S) is the smallest subspace of V containing S, in the following sense: if W is any
subspace of V with S ⊆ W , then span(S) ⊆ W .
Proof. The case when S = ∅ is left as an exercise. So assume S ̸= ∅.
(1) First show span(S) is closed under scalar multiplication. If x ∈ span(S) then (whether S is finite or
infinite) there exist u1 , . . . , un ∈ S and scalars a1 , . . . , an ∈ F with
x = a1 u1 , . . . , an un .
For any scalar c ∈ F we then get
c(a1 v1 + · · · + an vn ) = (ca1 )v1 + · · · + (can )vn
which is also a linear combination of u1 , . . . , un . This proves cx ∈ span(S) as required.
Next show span(S) is closed under addition. Let x, y ∈ span(S). By the earlier technical remark, there
exist v1 , . . . , vn ∈ S such that both x and y can be written as linear combinations of v1 , . . . , vn , say
x = a1 v1 + · · · + an vn
y = b1 v1 + · · · + bn vn .
Then
x + y = (a1 v1 + · · · + an vn ) + (b1 v1 + · · · + bn vn )
= (a1 v1 + b1 v1 ) + · · · + (an vn + bn vn ) (commutativity and associativity)
= (a1 + b1 )v1 + · · · + (an + bn )vn (VS 8).
This last expression is a linear combination of v1 , . . . , vn , vectors in S. This proves x + y ∈ span(S).
(2) If S = {v1 , . . . , vn }, then v ∈ S implies v = vi for some i. Then
v = 0·v1 + · · · + 0·vi−1 + 1·vi + 0·vi+1 + · · · + 0·vn
which proves v ∈ span(S) in this case. If S is infinite, then we can write each v ∈ S as v = 1·v, which is a
linear combination of finitely many (i.e., one) vector from S. So again v ∈ span(S).
(3) This will be proved on Wednesday. □

2
MATH 146 January 18 Section 2

Theorem 1.6. Let V be vector space over F. Let S ⊆ V . Then:

(1) span(S) is a subspace of V .
(2) span(S) contains S, i.e., S ⊆ span(S).
(3) Moreover, span(S) is the smallest subspace of V containing S, in the following sense: if W is any
subspace of V with S ⊆ W , then span(S) ⊆ W .
Proof (continued). (3) Suppose W is a subspace of V and S ⊆ W . Let x ∈ span(S). So x = a1 v1 +· · ·+an vn
for some v1 , . . . , vn ∈ S and a1 , . . . , an ∈ F. Then
• v1 , . . . , vn ∈ W (as S ⊆ W )
• So a1 v1 , a2 v2 , . . . , an vn ∈ W (as W is a subspace, so is closed under scalar multiplication)
• So a1 v1 + · · · + an vn ∈ W (as W is closed under addition)
I.e., x ∈ W . Since x was arbitrary in span(S), this proves span(S) ⊆ W . □
Comment. Theorem 1.6 gives an easy way to prove span(S) ⊆ W when S ⊆ V and W is a subspace of
V : just prove S ⊆ W and then use Theorem 1.6(3) to deduce span(S) ⊆ W .
Jargon. If S ⊆ V and span(S) = W , then we say that S spans the subspace W , or that S is a spanning
set of W . In particular, if span(S) = V then we say S spans V .
A basic question we will be frequently faced with is deciding whether a given vector x is in the span of
a given set of vectors.
Example. Consider the vector space M2×2 (R) of 2 × 2 matrices over R. Determine whether
         
2 1 1 0 1 1 0 1 1 2
∈ span , , , .
−1 1 −1 0 0 −1 1 1 1 0
Solution. We are asking whether there exist scalars a, b, c, d ∈ R satisfying
         
1 0 1 1 0 1 1 2 2 1
a +b +c +d = .
−1 0 0 −1 1 1 1 0 −1 1
Equivalently,          
a 0 b b 0 c d 2d 2 1
+ + + =
−a 0 0 −b c c d 0 −1 1
or more simply    
a + b + d b + c + 2d 2 1
= .
−a + c + d −b + c −1 1
Since matrices are equal if and only if their corresponding entries are equal, this last equation simplifies to
the following system of four linear equations:


 a + b + d = 2
b + c + 2d = 1

(∗)

 −a + c + d = −1
 −b + c = 1
The point of this example is that the original question (is the first matrix in the span of the other four
matrices?) is equivalent to asking whether the above system of 4 linear equations in the four unknowns
a, b, c, d has a solution in real numbers (remember that in this example, the scalars are real numbers).
We determine whether the system has a solution by solving it. There are many ways to do this (software,
row-reduction of an appropriate matrix). For now, we want you to know how to solve the system by direct
manipulation of the equations, called the method of elimination. I’ll illustrate in this example.
Step 1: eliminate the variable a from the 3rd equation, by adding the first equation to it. This changes
the system to


 a + b + d = 2
b + c + 2d = 1


 b + c + 2d = 1
 −b + c = 1
It’s easy to argue that every solution to the first system is also a solution to the 2nd system. The
converse is also true, since we can return to the original system by subtracting the first equation from the
(new) third equation. In other words, the original system and the new system are equivalent (have the
same solutions).
The 3rd equation is now irrelevant since it is a repeat. So we can delete it.

 a + b + d = 2
b + c + 2d = 1
 −b + c = 1
Step 2. There are no other occurrences of a that we can eliminate, so we consider eliminating another
variable. For example, we could eliminate b from the first and second equations by adding the third
equation to each of them. This gives

 a + c + d = 3
+ 2c + 2d = 2
−b + c = 1


Again, this system is equivalent to the original one.

Step 3. An obvious simplification to do here is to divide the second equation by 2 (i.e., multiply by 21 ).
Let’s also multiply the 3rd equation by −1. These simplifications do not change the solution set of the
system.

 a + c + d = 3
c + d = 1
 b − c = −1
Step 4. Can we eliminate another variable? If I try to eliminate c from the 2nd equation, I would need to
either add the 3rd equation to it, but that would re-introduce b; or subtract the first equation from it, but
that would introduce a. So I don’t want to do that. However, I could add the 2nd equation to the third
and eliminate c from the 3rd equation; and I could subtract the 2nd equation from the first and eliminate
c from it. Let’s do that:

 a = 2
(∗∗) c + d = 1
b + d = 0


Step 5: Neither occurrence of d can be eliminated (without re-introducing a, b or c). This signals that we
are done with the elimination process. Since we didn’t find a contradiction (such as 0 = 1), this system
has solutions. Since solutions exist, this means that the answer to our original question is YES:
         
2 1 1 0 1 1 0 1 1 2
is in span , , , .
−1 1 −1 0 0 −1 1 1 1 0
To actually see how to express the first matrix as a linear combination of the other four, we first describe
all solutions to the system (∗). Introducing a new symbol, or parameter, for d, say d = t, we can express
2
a, b, c, d in terms of t using the equations in (∗∗), obtaining
a = 2
b = − t
t ∈ R.
c = 1 − t
d = t
As t ranges over R, the equations above describe all solutions to (∗). For example, when t = 0 we get the
solution (a, b, c, d) = (2, 0, 1, 0). Thus one way to describe the original matrix as a sum of the other four is
“2 times the first matrix plus the third matrix,” i.e.,
         
2 1 1 0 1 1 0 1 1 2
=2· +0· +1· +0· .
−1 1 −1 0 0 −1 1 1 1 0
More information about the method of elimination will be placed on Learn.

3
MATH 146 January 20 Section 2

Properties of span. Let V be a vector space and S, T ⊆ V .

(1) If S ⊆ T then span(S) ⊆ span(T ).
Proof. Can prove this directly. Or argue like this: T ⊆ span(T ) (Theorem 1.6), so S ⊆ T =⇒
S ⊆ span(T ) =⇒ span(S) ⊆ span(T ) (using Theorem 1.6, since span(T ) is a subspace). □
?
(2) span(S ∪ T ) = span(S) ∪ span(T )
Answer. No: right-hand side might not be a subspace; left-hand side is always a subspace. □

§1.5
Jargon. Let V be a vector space over F. Given a linear combination
a1 v1 + · · · + an vn
of vectors from V , we say call it trivial if a1 = a2 = · · · = an = 0, and nontrivial otherwise.
Definition. Let V be a vector space over a field F. Let S ⊆ V .
(1) Say S is linearly dependent if there exists a nontrivial linear combination of distinct vectors from
S which equals 0; i.e., if it is possible to write
(∗) a1 v1 + a2 v2 + · · · + an vn = 0
for some distinct vectors v1 , . . . , vn ∈ S , and such that at least one ai is ̸= 0 .
(2) Say S is linearly independent if S is not linearly dependent: i.e., if

v1 , . . . , vn ∈ S 
v1 , . . . , vn distinct =⇒ a1 = a2 = · · · = an = 0.
a1 v1 + · · · + an vn = 0 

Example. (1) In R2 , the set {(1, −1), (−2, 2), (3, 4)} is linearly dependent because we can write
       
1 −2 3 0
2 +1 +0 = .
−1 2 4 0
(2) In F3 , is the set {(1,
  1, 0), (1,
 0,1), (0,1, 
1)} linearly
  dependent or independent?
1 1 0 0
Solution: solve a 1 + b 0 + c 1 = 0. We get the system
0 1 1 0
a + b = 0
a + c = 0
b + c = 0.
If F = R (or C or Q), we can solve to get a = b = c = 0, meaning the set is linearly independent.
But if F = Z2 , we get the solution a = b = c = t (t ∈ Z2 ). In fact, the sum of the three vectors in
(Z2 )3 is the zero vector, so in (Z2 )3 these vectors are linearly dependent.
(3) In any vector space, {0} is linearly dependent. Any set containing 0 is linearly dependent.
(4) {v} is linearly dependent iff v = 0. (Use A1 Problem 3 for ⇒.)
(5) What about ∅? Is it linearly dependent, or independent? (Answer: independent)
 (6) Suppose S ⊆ T ⊆ V . Which of the following implications are correct?
?
S is linearly dependent =⇒ T is linearly dependent. (True)
?
T is linearly dependent =⇒ S is linearly dependent. (False)
?
S is linearly independent =⇒ T is linearly independent. (False)
?
T is linearly independent =⇒ S is linearly independent. (True)
Proposition. Let V be a vector space over F. Let S ⊆ V . TFAE:
(1) S is linearly dependent.
(2) S = {0} or ∃v ∈ S such that v can be expressed as a linear combination of other vectors in S.
Proof. (⇐) We have already explained why {0} is linearly dependent. Suppose the vector v ∈ S can be
written as a linear combination of other vectors u1 , . . . , un ∈ S, say
v = c1 u1 + · · · + cn un , c1 , . . . , cn ∈ F.
We can assume that u1 , . . . , un are distinct. Since v ̸∈ {u1 , . . . , un } by assumption, it follows that
u1 , . . . , un , v are distinct. Now note that (−1)v = −v by Theorem 1.2(3) and so
c1 u1 + · · · + cn un + (−1)v = 0.
As u1 , . . . , un , v are distinct vectors in S and at least one of the coefficients (−1) in the above linear
combination is not 0, we get that S is linearly dependent.
(⇒) Assume S is linearly dependent. By assumption there exist distinct u1 , . . . , un ∈ S and a1 , . . . , an ∈
F, not all 0, such that
(∗) a1 u1 + · · · + an un = 0.
By “weeding out” terms where ai = 0, we can assume that ai ̸= 0 for all i = 1, . . . , n.
Case 1: n = 1. Then a1 u1 = 0 with a1 ̸= 0. Thus u1 = 0 by A1 Problem 3, so {0} ⊆ S. One option is
S = {0}; the other is that there exists v ∈ S with v =
̸ 0. In the second option, we can write 0 ∈ S as a
linear combination of v ∈ S, namely, 0 = 0v.
Case 2: n > 1.
In this case we can show that every ui can be written as a linear combination of the other uj ’s. For
example, here is the proof that u1 can be written as a linear combination of u2 , . . . , un : rewite (∗) as
a1 u1 = (−a2 )u2 + · · · + (−an )un .
Recall that a1 ̸= 0, so a−1
1 exists (in F). Multiply both sides of the equation to get
u1 = (−a2 /a1 )u2 + · · · + (−an /a1 )un
proving u1 ∈ S can be written as a linear combination of u2 , . . . , un ∈ S. □
Comment. Another equivalent condition to “S is linearly dependent” is
∃v ∈ S with v ∈ span(S \ {v}).
Indeed, if v ∈ S and v = a1 u1 + · · · + an un with u1 , . . . , un ∈ S and ui ̸= v for all i, then v ∈ span(S \ {v}).
If S = {0} then 0 ∈ span(∅) = span(S \ {0}).
Taking the negation, we get
Corollary. A set S in a vector space is linearly independent iff ∀v ∈ S we have v ̸∈ span(S \ {v}).

2
MATH 146 January 23 Section 2

§1.6 Recall: a set S (in a vector space/F) is linearly independent if


v1 , . . . , vn ∈ S 
v1 , . . . , vn distinct =⇒ a1 = a2 = · · · = an = 0.
a1 v1 + · · · + an vn = 0 

Definition. Let V be a vector space over F. A subset S ⊆ V is a basis for V if S is linearly independent
and spans V .
Example. (1) V = Rn : a basis is En = {(1, 0, 0, . . .), (0, 1, 0, . . .), . . . , (0, . . . , 0, 1)}.
| {z } | {z } | {z }
e1 e2 en
(2) More generally, En is a basis for Fn (for any F).
(3) V = Mm×n (F): a basis is Em,n = {Eij : 1 ≤ i ≤ m, 1 ≤ j ≤ n} where Eij is the matrix of all 0’s
except one 1 in the (i, j) spot.
(4) V = P(F): a basis is B = {1, x, x2 , . . .}.
Proof sketch of (4). Show span(B) = P(F): ⊆ is obvious. For ⊇, let f ∈ P(F), say
f = a0 + a1 x + a2 x2 + · · · + an xn
= a0 ·1 + a1 x + · · · + an xn .
So f is a linear combination of 1, x, x2 , . . . , xn , all in B; so f ∈ span(B). This proves P(F) ⊆ span(B).
Prove B is linearly independent: assume xi1 , . . . , xik are distinct vectors in B, ai1 , . . . , aik ∈ F, and
ai1 xi1 + · · · + aik xik = 0.
By definition, a polynomial is the zero polynomial iff all of its coefficients are 0, so
ai 1 = · · · = ai k = 0
proving B is linearly independent. So B is a basis. □
(5) Here is another basis for P(R), invented by Legendre: {P0 , P1 , . . . , Pn , . . .} where
n    k
X n n+k x−1
Pn = .
k=0
k k 2
For example, P0 = 1, P1 = x, P2 = 12 (3x2 − 1), . . . This basis for P(R) is “better” than B in some
applications.
Theorem 1.8. Suppose B = {v1 , . . . , vn } ⊆ V is finite. TFAE:
(1) B is a basis for V .
(2) Every v ∈ V admits a unique representation v = a1 v1 + · · · + an vn as a linear combination of B.
Proof sketch. (1) ⇒ (2). Since B spans V , every vector in V admits some representation as a linear
combination of B. To prove uniquencess, assume that v ∈ V admits two representations
v = a1 v1 + · · · + an vn
and v = b1 v1 + · · · + bn vn .
Subtract the 2nd from the first to get
0 = (a1 − b1 )v1 + · · · + (an − bn )vn .
Use linear independence of B to get
a1 − b1 = · · · = an − bn = 0, i.e., ai = bi for all i = 1, . . . , n.
 (2) ⇒ (1). By assumption, every v ∈ V admits a representation as a linear combination of B, so
span(B) = V . To prove B is linear independent, note that one representation of the zero vector (as a
linear combination of B) is
0 = 0v1 + · · · + 0vn .
By the uniqueness assumption, this is the only representation of the zero vector. That is,
a1 v1 + · · · + an vn = 0 =⇒ a1 = · · · = an = 0,
proving B is linear independent. □
Corollary. Let V be a vector space/F. If V has a finite basis B = {v1 , . . . , vn }, then V is “naturally” in
1-1 correspondence with Fn via
a1 v1 + · · · + an vn ↭ (a1 , . . . , an )
| {z } | {z }
∈V ∈Fn

We’ll discuss this correspondence more later.

Existence of bases
Theorem 1.9. Suppose the vector space V is spanned by a finite set S. Then S can be “shrunk” to a
basis B of V (i.e., ∃ basis B ⊆ S).
Proof sketch. We have span(S) = V . If S is linearly independent, then S itself is a basis. Otherwise,
∃v ∈ S such that v ∈ span(S \ {v}), and hence span(S \ {v}) = span(S) = V . Choose such v and let
S ′ = S \ {v}. It still spans V but is smaller. Repeat until you can’t anymore; the final S (k) must be linearly
independent and a basis. □
Question: is the theorem true for infinite spanning sets? More care would be needed in the proof. For
example, it is possible to construct a vector space V ̸= {0} and an infinite spanning set {v0 , v1 , . . .} with
the property that for every k, vk ∈ span({vk+1 , vk+2 , . . .}). So you might first throw away v0 , then v1 , then
v2 , etc. and at the end you will have reduced S to ∅ (which doesn’t span V ).1

1And even more problems arise if S is uncountable.

2
MATH 146 January 25 Section 2

Question. Suppose V has a finite linearly independent set S. Can S always be “grown” (i.e., extended)
to a basis for V ?
Naive algorithm proving “YES.”
Assume S ⊆ V is linearly independent. If span(S) = V , then S is already a basis. Otherwise, there
exists x ∈ V with x 6∈ span(S). Let S 0 = S ∪ {x}. I want to say that S 0 is also linearly independent. (I’ll
justify this later.)
Repeat: if span(S 0 ) = V then S 0 is a basis. Otherwise, there exists x0 ∈ V with x0 6∈ span(S 0 ). Let
S 00 = S 0 ∪ {x0 }. Then S 00 is still linearly independent.
Continue: we get S ⊂ S 0 ⊂ S 00 ⊂ · · · ⊂ S (n) ⊂ · · · with each set S (k) linearly independent. Now there is
the question of whether this process terminates. It turns out that we can prove termination, provided we
assume that V has a finite spanning set.
To fully justify the claims above, we need the following two theorems.
Theorem 1.7. Suppose V is a vector space, S ⊆ V , and x ∈ V with x 6∈ S. Then
S ∪ {x} is linearly independent ⇐⇒ (S is linearly independent and x 6∈ span(S)).
Proof sketch. (⇒) By known facts.
(⇐) Assume S is linearly independent and x 6∈ span(S). Suppose S ∪ {x} is linearly dependent. So
there exist v1 , . . . , vn ∈ S ∪ {x} (distinct) and scalars a1 , . . . , an ∈ F, not all 0, with
(∗) a1 v1 + · · · + an vn = 0.
By weeding out those ai vi for which ai = 0, we can assume that we have ai 6= 0 for all i.
We must have x ∈ {v1 , . . . , vn } (why?) WLOG assume that vn = x. Since an 6= 0, we can “solve for x”
in (∗) to get x ∈ span(v1 , . . . , vn−1 ) ⊆ span(S), contradiction. 
Theorem 1.10 (Baby Replacement Theorem1). Suppose V is spanned by some finite set of size n. Then
every linearly independent set S ⊆ V satisfies |S| ≤ n.
(Proof deferred until Friday). 
Now consider again the Question and the naive algorithm for solving it. Assume V is spanned by a
finite set of size n, and S is linearly independent. Recall the sequence of growing linearly independent sets
S ⊂ S 0 ⊂ S 00 ⊂ · · ·
By the Baby Replacement Theorem, all of these sets must have size ≤ n, so the sequence cannot go on
forever. Therefore at some step, say S (k) , the algorithm terminates, which must be because S (k) is a basis.
This proves:
Corollary 2 (c). Suppose V is spanned by a finite set. Then every linearly independent set S ⊆ V can be
extended to a basis for V .
Here is another application of the Baby Replacement Theorem.
Corollary 1. Assume V is spanned by a finite set.
(1) V has a finite basis B.
(2) Let n = |B|. Every basis of V has size n.
1The full Replacement Theorem says that, if T is the spanning set of size n, then not only do we have |S| ≤ n, but also
there exists a subset H ⊆ T of size |H| = n − |S|, such that S ∪ H spans V .
Proof. (1) Let S be a finite spanning set. S can be shrunk to a basis B, by Theorem 1.9 (Jan 23). Obviously
B is also finite.
(2) Let C be another basis. Apply the Baby Replacement Theorem 1.10 using B as the finite spanning
set and S := C as the linearly independent set to get |C| ≤ |B|. So C is also a finite basis. So we can
reverse the roles of B and C in the above argument to get |B| ≤ |C|. So |C| = |B| = n. 

2
MATH 146 January 27 Section 2
Recap: suppose V has a finite spanning set.
• Theorem 1.9: finite spanning sets shrink to bases.
• (Baby) Replacement Theorem: finite spanning set B, S lin. indep. =⇒ |S| ≤ |B|.
• Corollary 2(c): lin. indep. sets extend to bases.
• Corollary 1(1): V has a basis.
• Corollary 1(2): all bases have same (finite) size.
Still need to prove the (Baby) Replacement Theorem. First we prove:
Exchange Lemma∗ . Suppose V is a vector space, span(x1 , . . . , xk , v1 , . . . , v` ) = V (possibly with k = 0),
and x ∈ V with x 6= span(x1 , . . . , xk ). Then there exists i ∈ {1, . . . , `} such that
span(x1 , . . . , xk , x, v1 , . . . , vi−1 , vi+1 , . . . , v` ) = V.

Proof sketch. Let S = {x1 , . . . , xk , v1 , . . . , v` }. Since span(S) = V we can write

(∗) x = a1 x1 + · · · + ak xk + b1 v1 + · · · + b` v` .
At least one bi 6= 0 (why?); choose such i. Let T = (S \ {vi }) ∪ {x}. Using bi 6= 0, we can manipulate (∗)
to show vi ∈ span(T ). Clearly S \ {vi } ⊆ T ⊆ span(T ), so
S ⊆ span(T )
so
span(S) ⊆ span(T )
(by Theorem 1.6), proving span(T ) = V . 
(Baby) Replacement Theorem. Suppose V = span(B) with |B| = n. Then every linearly independent
set S ⊆ V satisfies |S| ≤ n.
Proof. Write B = {v1 , . . . , vn }. Assume there is a counter-example; then there is one with |S| = n + 1.
Write S = {x1 , . . . , xn+1 }. Observe that
span(v1 , v2 , . . . , vn ) = V and x1 6∈ span(∅) (why?).
| {z }
B

So by the Exchange Lemma, there exists i ∈ {1, . . . , n} so that

B 0 := (B \ {vi }) ∪ {x1 } still spans V .
For concreteness, let’s assume it is i = 1: so
span(x1 , v2 , . . . , vn ) = V.
| {z }
B0

Repeating, we have x2 6∈ span(x1 ) (why?). Again by the Exchange Lemma, there exists j ∈ {2, . . . , n}
such that (B 0 \ {vj }) ∪ {x2 } still spans V . For concreteness, assume it is j = 2, so
span(x1 , x2 , v3 , . . . , vn ) = V.
Keep repeating until we run out of v’s. Then we will have
span(x1 , x2 , . . . , xn ) = V.
In particular, xn+1 ∈ span(x1 , . . . , xn ), contradicting that S is linearly independent. 
 Corollary 1 justifies the next definition.
Definition. Let V be a vector space.
(1) Suppose V is spanned by a finite set. The dimension of V , written dim V , is the unique size n of
every basis of V . We also say that V is finite-dimensional.
(2) If V has no finite spanning set, then we say V is infinite-dimensional.
Example.
(1) Fn . It has the finite basis En of size n. So dim Fn = n.
(2) Mm×n (F). It has the finite basis Em,n of size mn. So dim Mm×n (F) = mn.
(3) P(F). It has the infinite basis {1, x, x2 , . . .} (which is linearly independent). By the Replacement
Theorem, P(F) has no finite spanning set, i.e., is infinite-dimensional.
(4) F(R, R). It has an infinite linearly independent subset (A2). So by the Replacement Theorem, it
has no finite spanning set. So it is infinite-dimensional.
Here is an intuitively obvious but important consequence of our results.
Theorem 1.11. Let V be a finite-dimensional vector space. Let W be a subspace. Then:
(1) W is finite-dimensional and dim W ≤ dim V .
(2) If W is a proper subspace (i.e., W ⊂ V ), then dim W < dim V .
Proof. Let n = dim V . Fix a subspace W . By the Replacement Theorem or Corollary 2(c), every lin.
indep. set in V has size ≤ n. In particular, every lin. indep. set in W has size ≤ n. This means that if we
start with some lin. indep. subset S ⊆ W (say S = ∅), the Naive Algorithm for extending S to a basis
for W will terminate, say at C. Then |C| = dim W (by definition). As C is linearly independent (in W ,
and hence in V ), we have |C| ≤ n (see the comments at the start of this proof). Thus dim W ≤ n, proving
(1).
(2) Let C be the basis for W we found in (1). C is linearly independent in V , so we can extend C to a
basis B of V (by Corollary 2(c)). We have C ⊆ B, |C| = dim W ≤ n, and |B| = dim V = n. If C = B, then
we would have W = span(C) = span(B) = V , contradiction. So C ⊂ B, so |C| < |B|, proving (2). 
Here is another cute result that is sometimes useful.
Corollary 2 (continued). Assume V is a finite-dimensional vector space with dim V = n. Suppose S ⊆ V
is a subset with |S| = n. Then
(1)
S is linearly independent ⇐⇒ S spans V ⇐⇒ S is a basis for V .
(1)
Proof. (⇒) By Corollary 2(c), S can be extended to a basis C. Then |C| = n. But S ⊆ C and |S| = n. So
S = C and S is already a basis for V .
(1)
(⇐) By Theorem 1.9 (Jan 23), we can shrink S to a basis C. Then |C| = n. But C ⊆ S and |S| = n. So
S = C and S is already a basis for V . 

2
MATH 146 January 30 Section 2

The material in this week’s lectures is not in our textbooks (only in the exercises of Friedberg-Insel-Spence).
Example. V = M2×2 (R),
     
a b 0 b
W1 = | a, b, c ∈ R , W2 = | b, d ∈ R .
c 0 b d
Warm-up: what is dim W1 ? dim W2 ? W1 ∩ W2 ? dim(W1 ∩ W2 )?
Definition. Suppose V is a vector space and W1 , W2 are subspaces. Their sum is the set
def
W1 + W2 = {w1 + w2 : wi ∈ Wi for i = 1, 2} ⊆ V.
Example. V = M2×2 (R), W1 and W2 as in the previous lecture. W1 + W2 = V .
Example. V = R3 , W1 = {the x-axis} = span(e1 ) and W2 = {the y-axis} = span(e2 ). Then
W1 + W2 = {(a, 0, 0) + (0, b, 0) : a, b ∈ R} = {(a, b, 0) : a, b ∈ R} = {the x, y-plane} = span(e1 , e2 ).
Proposition 1. If W1 , W2 are subspaces of V , then W1 + W2 is a subspace of V . Moreover, W1 + W2
contains both W1 and W2 .
Proof. 0 = 0 + 0 ∈ W + W 0 so W + W 0 6= ∅.
Suppose x, y ∈ W1 + W2 (must show x + y ∈ W1 + W2 ). By definition this means x = w1 + w2 and
y = u1 + u2 for some w1 , u1 ∈ W1 and w2 , u2 ∈ W2 . So
x + y = (w1 + w2 ) + (u1 + u2 )
= (w1 + u1 ) + (w2 + u2 ) (by commutativity, associativity).
| {z } | {z }
∈W1 ∈W2

So x + y ∈ W1 + W2 . A similar proof, using one of the distributive axioms, shows W1 + W2 is closed under
scalar multiplication. So W1 + W2 is a subspace of V .
Finally, w1 ∈ W1 =⇒
w1 = w1 + 0 ∈ W1 + W2
so W1 ⊆ W1 + W2 . Similarly, w2 = 0 + w2 ∈ W1 + W2 shows W2 ⊆ W1 + W2 . 
Remark. W1 + W2 is the smallest subspace containing both W1 and W2 ; that is, every subspace that
contains W1 and W2 also contains W1 + W2 .
Proposition 2. span(S ∪ T ) = span(S) + span(T ).
Proof. (⊆) S ⊆ span(S) ⊆ span(S) + span(T ). Similarly, T ⊆ span(T ) ⊆ span(S) + span(T ). So
S ∪ T ⊆ span(S) + span(T ).
Since the RHS is a subspace (Prop. 1), we get
span(S ∪ T ) ⊆ span(S) + span(T )
by Theorem 1.6.
(⊇) Clearly span(S ∪ T ) is a subspace containing both span(S) and span(T ). Thus span(S ∪ T ) contains
the smallest subspace containing span(S) and span(T ), which by a Remark is span(S) + span(T ). 

Remark. Can define W1 + W2 + · · · + Wk similarly. It is always a subspace, and is the smallest subspace
containing all W1 , . . . , Wk . span(S1 ∪ · · · ∪ Sk ) = span(S1 ) + · · · + span(Sk ).
Definition. If W1 , W2 are subspace of V , then we say that V is the direct sum of W1 and W2 , and we
write V = W1 ⊕ W2 , provided:
(1) W1 + W2 = V .
(2) Every v ∈ V can be written uniquely as v = w1 + w2 with w1 ∈ W1 and w2 ∈ W2 . This means if
v ∈ V and v = w1 + w2 = w10 + w20 with wi , wi0 ∈ Wi for i = 1, 2, then wi = wi0 for i = 1, 2.
Remark. We view this as “decomposing” or “factoring” the space V . This is especially useful in analyzing
the structure of linear operators on vector spaces (MATH 245).
  
0 b
Example. Let V = M2×2 (R) and W1 be as before. Let W3 = | b ∈ R . Then V = W1 ⊕ W3 ,
0 b
 
a b
because every ∈ V can be written
c d
     
a b a b−d 0 d
= +
c d c 0 0 d
| {z } | {z }
∈W1 ∈W3

and this is the unique way to write it.

Remark. We can similarly define V = W1 ⊕ W2 ⊕ · · · ⊕ Wk to mean
(1) V = W1 + · · · + Wk , and
(2) Every v ∈ V can be written uniquely as v = w1 + · · · + wk with wi ∈ Wi for i = 1, . . . , k.
Example. {v1 , . . . , vn } is a basis for V iff
V = span(v1 ) ⊕ span(v2 ) ⊕ · · · ⊕ span(vn ).
Theorem 1. V = W1 ⊕ W2 ⊕ · · · ⊕ Wk iff
(1) V = W1 + W2 + · · · + Wk , and
(20 ) the only solution to
0 = w1 + w2 + · · · + wk with w1 ∈ W1 , w2 ∈ W2 , . . . , wk ∈ Wk
is w1 = w2 = · · · = wk = 0.
Proof. Just need to prove (2) ⇐⇒ (20 ).
(2) ⇒ (20 ). We can write 0 = 0 + · · · + 0 (0 ∈ Wi for i = 1, . . . , k). By (2), this is the unique way to
write 0 = w1 + · · · + wk with wi ∈ Wi for all i, which proves (20 ).
(20 ) ⇒ (2). Assume (20 ). Suppose v ∈ V can be written
v = w1 + · · · + wk
and v = w10 + · · · + wk0
where wi , wi0 ∈ Wi for i = 1, . . . , k. Then
0 = v − v = (w1 + · · · + wk ) − (w10 + · · · + wk0 )
= (w1 − w10 ) + · · · + (wk − wk0 ) .
| {z } | {z }
∈W1 ∈Wk
0
(2 ) implies wi − wi0 = 0 for i = 1, . . . , k, so wi = wi0 for all i, proving (2). 
Theorem 2. V = W1 ⊕ W2 iff
(1) V = W1 + W2 , and
(200 ) W1 ∩ W2 = {0}.
Proof sketch. Just need to prove (20 ) ⇐⇒ (200 ) (where k = 2 in (20 )). The trick is to note that if w1 ∈ W1
and w2 ∈ W2 , then 0 = w1 + w2 ⇐⇒ w2 = −w1 =⇒ w1 ∈ W1 ∩ W2 . 
2
MATH 146 February 1 Section 2

Given subspaces W1 , W2 of vector space V , we have defined

W1 + W2 := {w1 + w2 : wi ∈ Wi for i = 1.2}.
If V is fin. dim., then so are W1 , W2 , and W1 + W2 (Theorem 1.11).
What can we say about dim(W1 + W2 )? Does it equal dim(W1 ) + dim(W2 )?
Theorem 1. Suppose V is finite-dimensional and W1 , W2 are subspaces of V . Then
dim(W1 + W2 ) = dim W1 + dim W2 − dim(W1 ∩ W2 ).
Before proving, let’s look at some applications.
(1) Suppose W1 ∩ W2 = {0}. Then dim(W1 + W2 ) = dim(W1 ) + dim(W2 ).
(2) In particular, if V = W1 ⊕ W2 , then dim(V ) = dim W1 + dim W2 .
(3) Suppose dim V = 5 and dim W1 = dim W2 = 4. Then dim(W1 ∩ W2 ) ≥ 3.
Proof. We have
dim(W1 + W2 ) = 4 + 4 − dim(W1 ∩ W2 )
so
dim(W1 ∩ W2 ) = 8 − dim(W1 + W2 ).
But dim(W1 + W2 ) ≤ dim V = 5, so dim(W1 ∩ W2 ) ≥ 3. □
Proof of Theorem 1. Start by letting B = {v1 , . . . , vk } be a basis for W1 ∩ W2 . We can consider B as a
subset of W1 . Note that B is linearly independent. Hence by Corollary 2(c), B can be extended to a basis
C1 for W1 , say,
C1 = {v1 , . . . , vk , x1 , . . . , xℓ }.
Note: x1 , . . . , xℓ ∈ W1 \ W2 .
Similarly, B can be extended to a basis C2 for W2 , say,
C2 = {v1 , . . . , vk , y1 , . . . , ym }.
Now let
D = C1 ∪ C2 = {v1 , . . . , vk , x1 , . . . , xℓ , y1 , . . . , ym }.
Clearly |D| = k + ℓ + m. (This is where we use the observation that each xi ̸∈ W2 ; hence no xi can equal
any yj .) Clearly
span(D) = span(C1 ∪ C2 ) = span(C1 ) + span(C2 ) = W1 + W2 . □
Claim: D is linearly independent.
Proof of Claim. Assume
(∗) a1 v1 + · · · + ak vk + b1 x1 + · · · + bℓ xℓ + c1 y1 + · · · + cm ym = 0.
Then
(∗∗) a1 v1 + · · · + ak vk + b1 x1 + · · · + bℓ xℓ = −(c1 y1 + · · · + cm ym ) .
| {z } | {z }
∈W1 ∈W2

Thus both sides of (∗∗) are in W1 ∩ W2 . Since W1 ∩ W2 = span(B), we get

−(c1 y1 + · · · + cm ym ) = d1 v1 + · · · + dk vk for some d1 , . . . , dk ∈ F,
Then
c1 y1 + · · · + cm ym + d1 v1 + · · · + dk vk = 0.
| {z }
a linear combo of C2
C2 is linearly independent, so c1 = · · · = cm = 0. (Also d1 = · · · = dk = 0.)
So (∗) simplifies to
a1 v1 + · · · + ak vk + b1 x1 + · · · + bℓ xℓ = 0.
This is a linear combination of C1 , and C1 is linearly independent, so
a1 = · · · = ak = b1 = · · · = bℓ = 0.
So all the coefficients in (∗) are 0. This proves D is linearly independent.
So D is a basis for W1 + W2 , so dim(W1 + W2 ) = |D| = k + ℓ + m. On the other hand,
dim(W1 ) + dim(W2 ) − dim(W1 ∩ W2 ) = (k + ℓ) + (k + m) − k = k + ℓ + m.
as required. □

Next, we start quotient spaces.

Definition 1.1. Suppose V is a vector space and W is a subspace. Given v ∈ V , define
def
v + W = {v + w : w ∈ W }.
Example. Let V = R2 . Let W = span{(1, 1)}) = {(a, a) : a ∈ R}.

v+W W

Note that v + W = v ′ + W

v′

Jargon. v + W is called the translation of W by v, or the coset of W containing v. (Note that v ∈ v + W .)

Definition. Given a subspace W of V , we let
V /W = {v + W : v ∈ V }.
I.e., V /W is the set of all translations of W .

2
MATH 146 February 3 Section 2
Lemma 1. Let W be a subspace of V . Let u, v ∈ V . Then
u + W = v + W ⇐⇒ u − v ∈ W.
Proof. (⇒) Assume u + W = v + W . Observe that u ∈ U + W . So u ∈ v + W , meaning u = v + w for
some w ∈ W . Then u − v ∈ W .
(⇐) Assume u − v =: w ∈ W . Then
u + W = (v + w) + W = v + (w + W ) = v + W (as w + W = W ; exercise.) □
Lemma 2. Let W be a subspace of V . The translations of W partition V ; that is,
(1) Their union is V , and
(2) If u + W ̸= v + W then (u + W ) ∩ (v + W ) = ∅.
Proof. (1) is obvious (v ∈ v + W for each v ∈ V ). To prove (2), we prove its contrapositive. Assume
(u + W ) ∩ (v + W ) ̸= ∅. Choose x ∈ (u + W ) ∩ (v + W ). This means
x=u+w for some w ∈ W
′
=v+w for some w′ ∈ W .
Then u + w = v + w′ , so u − v = w′ − w ∈ W . So u + W = v + W by Lemma 1. □
Note: the translations of W are the equivalence classes of an equivalence relation we can define on V ,
called “congruence mod W .” The definition is
u≡v (mod W ) ⇐⇒ u − v ∈ W.

Recall that for a fixed subspace W , V /W denotes the set of all translations of W . Thus
elements of V /W ↭ (certain) subsets of V
We use vectors in V to “name” elements of V /W (v “names” v + W ), but the vector names are not the
translations. In particular, elements of V /W can be “named” in more than one way. For example, if
V = R2 , W = span({(1, 1)}), and S = (−1, 0) + W , then also S = (0, 1) + W , since
(−1, 0) − (0, 1) = (−1, −1) ∈ W, so (−1, 0) + W = (0, 1) + W (Lemma 1).
We want to turn the set V /W into a vector space. So we need to define addition and scalar multiplication
on the elements of V /W . The idea is illustrated in the following example.
Example. V = R2 , W = span({(1, 1)}), S = (−1, 0) + W , T = (2, −1) + W .
W

S S+T T

u = (−1, 0)

u+v v = (2, −1)

 def
In general, if S = u + W and T = v + W , then we want to define S + T = (u + v) + W . Is there a
problem if we used different names for S and T ?
Lemma (There is no problem). Suppose V is a vector space, W is a subspace, and u, u′ , v, v ′ ∈ V . Then
u + W = u′ + W

=⇒ (u + v) + W = (u′ + v ′ ) + W.
and v + W = v ′ + W
In other words, defining addition on V /W by
def
(u + W ) + (v + W ) = (u + v) + W
is well-defined (does not depend on the choice of names u, v).
Proof. The hypothesis implies u − u′ ∈ W and v − v ′ ∈ W . The conclusion will follow if we can establish
(u + v) − (u′ + v ′ ) ∈ W . Well,
(u + v) − (u′ + v ′ ) = (u − u′ ) + (v − v ′ ) ∈ W. □
| {z } | {z }
∈W ∈W

Remark. Similarly, if F is the field of scalars for V , then we can define scalar multiplication on V /W by
def
c · (v + W ) = (cv) + W
and there is no problem (it is well-defined: if v + W = v ′ + W then (cv) + W = (cv ′ ) + W ).
Proposition. V /W with the operations defined as above is a vector space over F.
Proof. Ugh. There are 8 axioms that we must verify. Some are easy: e.g., commutativity (VS 1). Suppose
S, T ∈ V /W , say S = x + W and T = y + W . Then
df ∗ df
S + T = (u + W ) + (v + W ) = (u + v) + W = (v + u) + W = (v + W ) + (u + W ) = T + S.
∗
(= marks an application of (VS 1) in V .) Or left distributivity (VS 7):
df
a(S + T ) = a((x + W ) + (y + W )) = a((x + y) + W )
df ∗
= (a(x + y) + W ) = ((ax + ay) + W )
df df
= (ax + W ) + (ay + W ) = a(x + W ) + a(y + W ) = aS + aT.
Existence of a zero vector is more interesting. What is the zero vector in V /W ? (It is W .)
And the additive inverse of v+W is (−v)+W , because obviously (v+W )+((−v)+W ) = 0+W = W . □
Now that we know that V /W is a vector space, we can talk about linear combinations of “vectors” (i.e.,
translations of W ), linear independence, span, etc. The next theorem tells us what is dim(V /W ) when
dim V is finite.
Theorem. Suppose If V is finite-dimensional, say dim V = n. Let W be a subspace, say of dimension k.
Then dim(V /W ) = n − k = dim V − dim W .
Proof. Let B = {v1 , . . . , vk } be a basis for W . We can view B as a linearly independent subset of V . So
we can extend B to a basis C = {v1 , . . . , vk , vk+1 , . . . , vn } for V . Let D := {vk+1 + W, . . . , vn + W }. We
will show that D is a basis for V /W . Since |D| = n − k, this will prove the theorem.
We first show that D is linearly independent (in V /W ). Suppose ak+1 , . . . , an ∈ F and
(∗) ak+1 (vk+1 + W ) + · · · + an (vn + W ) = W.
Using the definitions of the operations in V /W , we can simplify this to
(ak+1 vk+1 + · · · + an vn ) + W = W.
2
By Lemma 1, this implies
ak+1 vk+1 + · · · + an vn ∈ W,
so
ak+1 vk+1 + · · · + an vn = b1 v1 + · · · bk vk for some b1 , . . . , bk ∈ F,
since {v1 , . . . , v)k } spans W . Then
(−b1 )v1 + · · · + (−bk )vk + ak+1 vk+1 + · · · + an vn = 0.
This is a linear combination of C. Since C is linearly independent, all the coefficients are 0. In particular,
ak+1 = · · · = an = 0. This proves that the only solution to (∗) is the trivial one, so D is linearly
independent.
Next we will show that D spans V /W . Let v + W ∈ V /W be given. Because C spans V , we can write
v = a1 v1 + · · · + ak vk + ak+1 vk+1 + · · · + an vn for some a1 , . . . , an ∈ F.
Claim. v + W = ak+1 (vk+1 + W ) + · · · + an (vn + W ) (so v + W ∈ span(D)).
Proof of Claim. The RHS of the Claim simplifies to u + W where u = ak+1 vk+1 + · · · + an vn . We want to
prove v + W = u + W . We can use Lemma 1: it is enough to prove v − u ∈ W . Well,
v − u = (a1 v1 + · · · + ak vk + ak+1 vk+1 + · · · + an vn ) − (ak+1 vk+1 + · · · + an vn )
= a1 v1 + · · · + ak vk ,
which is in W because v1 , . . . , vk ∈ W and W is a subspace. □

3
MATH 146 February 6 Section 2

§2.1. Here is the definition of “good” functions between vector spaces.

Definition. Let V and W be vector spaces over the same field F. A function T : V → W is called a
linear transformation, or is said to be linear, if:
(1) T (x + y) = T (x) + T (y) for all x, y ∈ V , and
(2) T (ax) = aT (x) for all x ∈ V and a ∈ F.
A linear transformation that is also a bijection is called an isomorphism.
Remark: We often write T x in place of T (x).
Example.
(1) (F = R). Let V = W = R2 . Define T : R2 → R2 by T (x1 , x2 ) = (−x2 , x1 ). This maps, e.g.,
(1, 0) 7→ (0, 1) and (0, 1) 7→ (−1, 0).
In fact, this T is just “rotation c.c.w. by 90◦ about (0, 0).”
Claim. T is a linear transformation.
Proof. For any x = (x1 , x2 ), y = (y1 , y2 ) ∈ R2 ,
T (x + y) = T ((x1 , x2 ) + (y1 , y2 ))
= T ((x1 + y1 , x2 + y2 )) (def. of + in R2 )
= (−(x2 + y2 ), x1 + y1 ) (def. of T )
= (−x2 , x1 ) + (−y2 , y1 )
= T (x) + T (y)
and similarly,
T (ax) = T (a(x1 , x2 ))
= T ((ax1 , ax2 ))
= (−ax2 , ax1 )
= a(−x2 , x1 )
= aT (x). □
In fact, T is an isomorphism (from R to R).
(2) V = W = R2 , T : R2 → R2 given by T (x1 , x2 ) = ( x1 +x
2
2 x1 +x2
, 2 ). The range of T is the x, y-line. In
2
effect, T maps each point in R to the closest point on this line.
Claim. T is linear.
Proof. Exercise. □
(3) The previous two examples are special cases of the following: let F be a field, and let A ∈ Mm×n (F),
say
a11 a12 · · · a1n
 
 a21 a22 · · · a2n 
A =   ... ... .. 
. 
am1 am2 · · · amn
 Given x = (x1 , . . . , xn ) ∈ Fn , define
a11 a12 · · · a1n x1 a11 x1 + a12 x2 + · · · + a1n xn
    
 a21 a22 · · · a2n   x2  def  a21 x1 + a22 x2 + · · · + a2n xn 
Ax =  =   ∈ Fm .
 ... .. ..  ..  ..

. .  .   . 
am1 am2 · · · amn xn am1 x1 + am2 x2 + · · · + amn xn
(This is “matrix-vector multiplication.”) Note that the i-th entry of the product Ax is obtained by
combining the entries in the i-th row
 of A with
 the entries of x “like a dot product.”
0 2 3
For example, if F = Z5 and A = ∈ M2×3 (Z5 ) and x = (3, 2, 1) ∈ (Z5 )3 , then
4 1 2
 
  3    
0 2 3   0·3 + 2·2 + 3·1 2
Ax = 2 = = ∈ (Z5 )2 .
4 1 2 4·3 + 1·2 + 2·1 1
1
Jeff ’s comment: Ax can also be defined as the linear combination of the columns of A using the
entries of x as scalars. So in the above example,
 
  3              
0 2 3   0 2 3 0 4 3 2
Ax = 2 =3 +2 +1 = + + = .
4 1 2 4 1 2 2 2 2 1
1
(Thank you, Jeff!) Returning to the general situation: given A ∈ Mm×n (F), define LA : Fn → Fm
by LA (x) = Ax.
Claim. Every such map LA : Fn → Fm is linear.
Proof. Tedious but not difficult. □
 
0 −1
Note: in Example 1 we had T = LA : R2 → R2 with A = . In example 2 we had
1 0
 
2 2 1/2 1/2
T = LA : R → R with A = .
1/2 1/2
(4) Define D : P(R) → P(R) by D(f ) = f ′ (derivative). D is linear. Is it injective? Surjective?
(5) Let C(R) be the set of all continuous functions R → R. Define T : C(R) → C(R) by
Z x
(T f )(x) = f (t)dt.
0
T is linear. Is it injective? Surjective?
Facts. Suppose V, W are vector spaces and T : V → W is a function.
(1) If T is linear, then
• T (0) = 0. (Proof: T (0) = T (0 + 0) = T (0) + T (0).)
• T (−x) = −T (x). (Hint: −x = (−1)x.)
• T (a1 x1 + · · · + an xn ) = a1 T (x1 ) + · · · + an T (xn ). (By induction on n)
(2) T is linear ⇐⇒ T (ax + by) = aT (x) + bT (y) for all x, y ∈ V and all a, b ∈ F. (Exercise)
Theorem 2.6. Let V, W vector spaces over F with V finite-dimensional. Suppose {v1 , . . . , vn } is a basis
for V . Let w1 , . . . , wn be any vectors in W (not necessarily distinct). Then there exists a unique linear
transformation T : V → W satisfying
T vi = wi , for i = 1, . . . , n.
2
Proof of Theorem 2.6. To prove existence, define T : V → W by
df
T (a1 v1 + · · · + an vn ) = a1 w1 + · · · + an wn .
| {z }
arbitrary element of V

Every element of V can be written as on the LHS, and in just one way (Thm. 1.8, Jan 23), so there is no
issue of being well-defined.
Clearly T vi = wi for each i = 1, . . . , n since
df
T vi = T (0·v1 + · · · + 1·vi + · · · + 0·vn ) = 0·w1 + · · · + 1·wi + · · · + 0·wn = wi .
Check that T is linear. Let x, y ∈ V . Say x = ni=1 ai vi and y = ni=1 bi vi . Then
P P

Xn Xn n
X n
X
T (x + y) = T ( (ai + bi )vi ) = (ai + bi )wi = ai w i + bi wi = T x + T y.
i=1 i=1 i=1 i=1

A similar proof shows T (cx) = cT (x).

To prove uniqueness, assume more generally that T, T ′ are two linear transformations V → W satisfying
T vi = T ′ vi for i = 1, . . . , n. We must show T = T ′ , i.e., T x = T ′ x for all x ∈ V . Fix x ∈ V ; because
{v1 , . . . , vn } spans V , we can write x = a1 v1 + · · · + an vn . Then
T x = T (a1 v1 + · · · + an vn )
= a1 T v1 + · · · + an T vn T is linear
= a1 T ′ v1 + · · · + an T ′ vn assumption
′
= T (a1 v1 + · · · + an vn ) T ′ is linear
= T ′ x.
So T = T ′ as claimed. □

3
MATH 146 February 8 Section 2

Definition. Suppose T : V → W is linear. The range (or image) of T is

def
Ran(T ) = {T (x) : x ∈ V }.
Note: Ran(T ) ⊆ W . Text R(T ) in place of Ran(T ).
Theorem 2.1 (a). Let T : V → W be linear. Then Ran(T ) is a subspace of W .
Proof. Must show Ran(T ) 6= ∅, Ran(T ) is closed under +, and Ran(T ) is closed under scalar multiplication.
(1) T (0V ) = 0W so 0W ∈ Ran(T ). (Alternate proof: T is a function with nonempty domain, so duh.)
(2) Given T (x), T (y) ∈ Ran(T ), we have T (x) + T (y) = T (?), proving T (x) + T (y) ∈ Ran(T ).
(3) Given T (x) ∈ Ran(T ) and a ∈ F, we have aT (x) = T (?), proving aT (x) ∈ Ran(T ). 
The following will be useful.
Theorem 2.2. Suppose T : V → W is linear with V finite-dimensional. If V = span(v1 , . . . , vn ), then
Ran(T ) = span(T v1 , . . . , T vn ).
Proof. Ran(T ) = {T x : x ∈ V } = {T (a1 v1 + · · · + an vn ) : ai ∈ F}
= {a1 T v1 + · · · + an T vn : ai ∈ F} = span(T v1 , . . . , T vn ). 
Corollary. Suppose T : V → W is linear with V finite-dimensional. Then Ran(T ) is also finite-
dimensional and dim(R(T )) ≤ dim(V ).
Proof. Let n = dim V . By picking a basis for V , Theorem 2.2 gives a spanning set for Ran(T ) of size ≤ n.
We can shrink it to a basis for Ran(T ), which will also have size ≤ n. 
  
1/2 1/2 x1
Example. V = W = R2 , T : V → W given by T (x1 , x2 ) = ( x1 +x 2 x1 +x2
, 2 )= .
2 1/2 1/2 x2
   
1 0
We know that V is spanned by {e1 , e2 } = , . Hence by Theorem 2.2,
0 1
       
1/2 1/2 1/1 1
Ran(T ) = span(T e1 , T e2 ) = span , = span = span .
1/2 1/2 1/2 1
Definition. Suppose T : V → W is linear. The kernel (or null space) of T is
def
Ker(T ) = {x ∈ V : T x = 0W }.
Note: Ker(T ) ⊆ V . Text uses N (T ) in place of Ker(T ).
Example. Same T : R2 → R2 .
       
2 x1 x1 +x2 x1 +x2 x1 −1
Ker(T ) = {x ∈ R : T x = 0} = :( 2 , 2 )=0 = : x1 + x2 = 0 = span .
x2 x2 1
Theorem 2.1 (b). Let V be linear. Then Ker(T ) is a subspace of V .
Proof. Again must show Ker(T ) is nonempty and closed under the vector space operations.
(1) T (0V ) = 0W , so 0V ∈ Ker(T ), so Ker(T ) 6= ∅.
(2) Suppose x, y ∈ Ker(T ). So T x = T y = 0. So T (x + y) = T x + T y = 0 + 0 = 0. So x + y ∈ Ker(T ).
This proves Ker(T ) is closed under addition.
(3) Suppose x ∈ Ker(T ) and a ∈ F. So T x = 0. So T (ax) = aT (x) = a0 = 0, so ax ∈ Ker(T ). This
proves Ker(T ) is closed under scalar multiplication. 
Coverage of midterm ends here
Theorem 2.4. Suppose T : V → W is linear. Then T is injective ⇐⇒ Ker(T ) = {0}.
Proof. (⇒) Assume T is injective. Since {0V } ⊆ Ker(T ), it suffices to show Ker(T ) ⊆ {0V }. Let x ∈
Ker(T ). So T x = 0W = T (0V ). By injectivity, we get x = 0V . This proves Ker(T ) ⊆ {0V }.
(⇐) Assume Ker(T ) = {0V }. Let x, y ∈ V and assume T x = T y. Then
T (x − y) = T (x + (−y)) = T x + T (−y) = T x + −T (y) = T x − T y = 0 (since T x = T y).
Hence x − y ∈ Ker(T ). But Ker(T ) = {0} so x − y = 0, so x = y. This proves T is injective. 
Theorem 2.3 (Rank-Nullity Theorem). Suppose T : V → W is linear with V finite-dimensional. Then
dim(Ran(T )) + dim(Ker(T )) = dim(V ).
| {z } | {z }
“rank of T ” “nullity of T ”