Math 108a Professor: Padraic Bartlett

Lecture 5: Span and Linear Independence

Week 2 UCSB 2013

Our lectures thus far have focused on the two concepts of fields and vector spaces.
In specific, we’ve studied these two objects in a fairly formal setting: we’ve created sets of
axioms that describe these objects, used these axioms to deduce certain properties that all
fields and vector spaces must possess, and built up a collection of useful examples of both
fields (R, Q, C, Z/2Z) and vector spaces (Rn , Cn , R[x], MR (n, n), and a bunch of subspaces
of these spaces.)
In this next sequence of lectures, we’re going to change our focus somewhat. Now that
we’ve built up a good set of examples, the natural thing to ask is what properties set them
apart? What makes them interesting? What sorts of properties can we study for each
individual space?
We study two such objects in this talk: the concepts of span and linear independence.

1 Span and Linear Independence

When we’re working with vectors in a vector space, there’s really only two operations we’ve
discussed that we can do: vector addition and scalar multiplication. A useful thing to think
about, then, is the collection of all sorts of objects that we can make using these operations!
We define these things here:
Definition. Let V be a vector space over some field F . A linear combination of some
set of vectors S is any sum of the form

a1 v~1 + a2 v~2 + . . . an v~n ,

where a1 , . . . an are all elements of our field F , and v~1 , . . . v~n ∈ S.

In other words, a linear combination of some set of vectors is anything we can make by
scaling and adding these vectors together.
A useful thing to study, given some set {v~1 , . . . v~n of vectors, is the following: what can we
make with these vectors? In other words, what is the collection of all linear combinations
of these vectors?
This question is sufficiently common that we have a term for its related concept: the
idea of span!
Definition. Given any collection of vectors A, the span of A, denoted span(A), is the
collection of all linear combinations of elements of A. In other words,

span(A) = {a1 v~1 + a2 v~2 + . . . an v~n |n ∈ N, a1 . . . an ∈ R, v~1 , . . . v~n ∈ A}.

A genre of questions you’ll often encounter in a linear algebra class is “Given some set
A of vectors, what is their span? Is some other vector w
~ contained within this span?”
We give an example of a calculation here:

1
Question. You have been hired as a deep-sea navigator. Your first task is to pilot a
submarine from our ship, anchored at (0, 0, 0) in the diagram below, to observe a whale
at (12, 6, −6). Your submarine has three engines on it. The first, when ran for a cycle,
moves the submarine (2, 1, 0) units from its current location. The second, when ran for a
cycle, moves the submarine (0, 2, 1) units from its current location. Finally, the third, when
ran for a cycle, moves the submarine (1, 0, 2) units from its current location. Submarine
engines can be fired for fractional units of time, and when ran in reverse moves the shuttle
backwards along that given vector.
z

x y

(12,6,-6)

Can you make it to the whale?

Answer. Essentially, our question is asking if the vector (12, 6, −6) is contained in the span
of the three vectors (2, 1, 0), (0, 2, 1), (1, 0, 2).
I claim that it is! To see why, simply just start trying to combine these three vectors
into (12, 6, −6). If we assume that we fire the first engine for a units, the second for b units,
and the third for c units, we’re essentially trying to find a, b, c such that

a(2, 1, 0) + b(0, 2, 1) + c(1, 0, 2) = (12, 6, −6).

This gives us three equations, one for the x-coördinate, one for the y-coördinate, and a third
for the z-coördinate:

2a + 0b + 1c = 12,
1a + 2b + 0c = 6,
0a + 1b + 2c = −6.

Subtracting two copies of the second equation from the first gives us

−4b + c = 0,

2
in other words c = 4b. Plugging this into the last equation gives us
b + 2(4b) = −6,
i.e. b = − 23 . This gives us then that c = − 83 , and thus that
8
2a − = 12,
3
i.e. a = 22
3 .
Consequently, we’ve just calculated that
22 2 8
(2, 1, 0) − (0, 2, 1) − (1, 0, 2) = (12, 6, −6);
3 3 3
in other words, that (12, 6, −6) is in the span of our set of vectors, and therefore that we
can get to it by using our three engines as described by a, b, c!
A fact worth noting is the following:
Proposition. Let V be a vector space over a field F and A any nonempty subset of vectors
of V . The span of A, span(A), is a subspace of V .
Proof. As on Friday, we just have to check three properties:
• Closure(+): Take any two vectors ~x, ~y ∈ span(A). For each of these vectors, be-
cause they’re in the span of A, we can write them as some linear combination of
elements of A: i.e. we can find field elements a1 , . . . an , b1 , . . . bm ∈ F and vectors
v~1 , . . . v~n , w~1 , . . . w~m ∈ A such that
a1 v~1 + . . . an v~n = ~x,
b1 w~1 + . . . bn w~n = ~y .
Then, we have that
a1 v~1 + . . . an v~n + b1 w~1 + . . . bn w~n = ~x + ~y .

In other words, ~x + ~y is a linear combination of elements in A! Therefore, ~x + ~y is

also in the span of A.
• Closure(·): Take any vector ~x ∈ span(A). Like before, because ~x is in the span of
A, we can write it as some linear combination of elements of A: i.e. we can find field
elements a1 , . . . an ∈ F and vectors v~1 , . . . v~n ∈ A such that
a1 v~1 + . . . an v~n = ~x.

Pick any b ∈ F . We have that

b~x = b (a1 v~1 + . . . an v~n ) = ba1 v~1 + . . . ban v~n ;

in other words, b~x is a linear combination of elements of A, and is therefore in the

span of A.

3
 • Identity(+). Take any vector ~v ∈ A. We know/have proven that 0~v = ~0, the additive
identity, for any vector space V . Therefore, because span(A) is closed under scalar
multiplication, we have just shown that ~0 is in span(A).

Because these three properties hold, by our results on Friday (which said that given these
three properties along with the fact that span(A) is a subset of V , the rest of the vector
space properties must follow) we have just shown that this is a subspace of V !

Based off of this observation, we can make the following definition:

Definition. Suppose that A = {~v1 , . . . ~vn } is some set of vectors, drawn from some vector
space V . In the proposition above, we proved that span(A) is a subspace of the vector
space V : in other words, span(A) is a vector space in its own right! Suppose that span(A)
is equal to the vector space U . Then we say that A spans U .

This definition motivates a second kind of question: take some vector space V . Can we
find a set A of vectors that spans V ?
The answer here is clearly yes: we can just pick A to be V itself! Then the span of A
is certainly V , because every vector of V is simply contained in A. Yet, this answer is also
kind of dumb. While A spans V , it does so with a lot of redundancy: i.e. if V was R3 , we’d
be using a set with infinitely many elements to span R3 , when we really only need the three
vectors (1, 0, 0), (0, 1, 0), (0, 0, 1).
Here’s a related question, to help us spot this kind of inefficiency: suppose we have some
set A of vectors. Is it possible to remove a vector from A and still have a set with the same
span as A?
This concept also comes up all the time in mathematics! Therefore, we have a definition
for it:

Definition. Let V be some vector space over a field F , and A be a subset of V . We say that
the set A is linearly dependent if there is some n > 0 and distinct elements v~1 , . . . v~n ∈ A,
field coefficients a1 , . . . an 6= 0 ∈ F such that

~0 = a1 v~1 + a2 v~2 + . . . an v~n .

In other words, a set A is linearly dependent if there is a linear combination of elements in

A that sums to 0.
If no such combination exists, then we say that A is linearly independent.

Notice that if a set is linearly dependent, then there is some vector within it that we
can remove without changing its span! We prove this here:

Lemma 1. Let V be a vector space over a field F , and S be a linearly dependent subset of
V . Then there is some vector ~v in S that we can remove from S without changing its span.

Proof. By definition, because S is linearly dependent, there is some n > 0 and distinct
elements v~1 , . . . v~n ∈ A, field coefficients a1 , . . . an 6= 0 ∈ F such that

~0 = a1 v~1 + a2 v~2 + . . . an v~n .

4
Solve for v~1 :
a2 a3 an
v~1 = − v~2 − v~2 − . . . v~n .
a1 a1 a1
This is a way to create the vector v~1 without using the vector v~1 itself.
Consider the set S \ {v~1 }, the set created by removing v~1 from S. We claim that this
set has the same span as S itself. To see why, take any element ~u in the span of S: i.e.
any linear combination of elements in S. Pick distinct vectors w~1 , . . . w~m ∈ S and scalars
b1 , . . . bm such that

~u = b1 w~1 + . . . bn w~n .

If none of these vectors w~1 , . . . w~n are the vector v~1 , then this linear combination is also in
the span of S \ {v~1 }. Otherwise, if some w ~i = v~1 , then we can just replace w~i with
a2 a3 an
− v~2 − v~2 − . . . v~n .
a1 a1 a1
This gives us a way to combine elements of S \ {v~1 } in a way to create ~u.
So, we’ve shown that any element of the span of S is also in the span of S \ {v~1 }!
Therefore, these two sets have the same span, and we’ve proven our claim: there is a vector,
v~1 , that we can remove from S without changing its span.

By repeatedly applying Lemma 1, we can get the following theorem:

Theorem 2. Any finite set of vectors S has a linearly independent subset T , such that
span(S) = span(T ).

Proof. Take any set S. If it is linearly independent, stop; we’re done. Otherwise, repeatedly
apply Lemma 1 to find an element we can remove from S without changing its span, and
remove that element from S. Because our set is finite, and any set with one vector is linearly
independent (prove this if you don’t see why!), we know that this process will eventually
stop and leave us with a linearly independent set at some stage!
So we’ve found a subset T of S with the same span as S, that’s linearly independent.
Yay!

Side note. In the infinite case, things are much more complicated. In particular, the
above process, of repeatedly removing elements one-by-one, isn’t guaranteed to ever stop!
For example, suppose you had the set S = {(z, 0) : z 6= 0 ∈ Z}.
This is an infinite set. Moreover, because the span of this set is simply {(x, 0) : x ∈ R},
literally any single element (z, 0) of S will have the same span as S: this is because xz (z, 0)
is a linear combination using just (z, 0) that can generate any element in span(S).
However, suppose you were simply removing elements one-by-one from S. It is possible,
through sheer bad luck, you would pick the elements (1, 0), (2, 0), (3, 0), (4, 0), . . . all in a
row. This would never get you down to a linearly independent set! In particular, you’d still
have all of the elements (z, 0) where z is a negative integer lying around, and there are tons
of linear combinations of these elements that combine to (0, 0).

5
 So the argument above doesn’t work. However, the result is still true: there is a subset
that is linearly independent with the same span! Proving that this always exists, though,
is tricky, and requires the use of things like the axiom of choice, which is beyond the scope
of this class.

To illustrate how we concretely find linearly independent subsets of certain sets of vec-
tors, we work an example:
Question. Consider the set

S = {a, a + 1, a + 2) : a ∈ N, 0 < a ≤ 100}.

Is this set linearly dependent? If it is, find a linearly independent subset of this set. If not,
prove it is linearly independent.
Proof. So, it certainly seems like this set should be linearly dependent: there are a hundred
vectors in the set, and it’s a subset of R3 ! However, we need to actually prove this is true,
which is a little tricker: where do we start?
One good tactic here is to look at some concrete elements of our set, and see what we
can do with those elements. (In general, this is a useful tactic when you’re presented with
a large collection of objects all described using rules: make a few examples and see what
those examples do!)
In particular, we know that (1, 2, 3), (2, 3, 4), (3, 4, 5), and (4, 5, 6) are all elements of our
set. Is this smaller, easier-to-understand set linearly dependent?
Well: if it were, we would have to have some way of combining elements to get to (0, 0, 0).
It’s not obvious how to do this as written, though. One strategy here is to just algebra-bash
out a solution: i.e. try to find a, b, c, d such that a(1, 2, 3) + b(2, 3, 4) + c(3, 4, 5) + d(4, 5, 6) =
(0, 0, 0). This will work! But it seems tedious; solving three equations in four variables
is not the most enjoyable thing in the world to do. Also, we’ve already given an example
(earlier, when we were showing that a given vector is in the span of some set) that basically
goes through this method.
An alternate strategy, that is sometimes worth pursuing, is not to just try to find
combinations of (1, 2, 3), (2, 3, 4), (3, 4, 5), (4, 5, 6) that get to (0, 0, 0), but rather to just
look for combinations that get to something useful! I.e. if we could make the vectors
(1, 0, 0) or (0, 1, 0) or (0, 0, 1), that would be really useful for making (0, 0, 0) later, as these
are relatively useful and easy-to-understand vectors!
This, we can do. In particular, notice that

(2, 3, 4) − (1, 2, 3) = (1, 1, 1),

which certainly looks like a useful vector!

In fact, it’s great for generating the other vectors in this set! Notice that, for example,

(3, 4, 5) = (1, 2, 3) + 2(1, 1, 1) = (1, 2, 3) + 2((2, 3, 4) − (1, 2, 3)),

and therefore that we can write (3, 4, 5) as a linear combination of vectors in our set!
More generally, notice that

(a, a + 1, a + 2) = (1, 2, 3) + (a − 1)(1, 1, 1) = (1, 2, 3) + (a − 1)((2, 3, 4) − (1, 2, 3)),

6
and therefore that we can write any vector in our set as a linear combination of the two
vectors (1, 2, 3) and (2, 3, 4)!
Therefore, we’ve shown that the span of our set S is just the span of the two vectors
(1, 2, 3) and (2, 3, 4). These vectors are clearly linearly independent: to see why, notice that
if

x(1, 2, 3) + y(2, 3, 4) = (0, 0, 0),

we would have to have

x + 2y = 0,
2x + 3y = 0, and
3x + 4y = 0.

The first two equations tell us that x = −2y and that x = − 23 y, which can only hold if
x = y = 0. Therefore, there is no nontrivial combination (i.e. no combination in which the
coefficients are nonzero) of these two vectors that is equal to (0, 0, 0).
So we’ve created a linearly independent subset of S that has the same span as S! Yay,
success.