Web Analytics
MTH 215 Chapter 4

MTH 215 — Intro to Linear Algebra

Kyle Monette
Spring 2026

Section 4.1: Vector Spaces and Subspaces

Many concepts regarding vectors in $\R^n $ can be extended to other mathematical systems. In this chapter we discuss collections of objects that behave like vectors do in $\R^n$, which we call vector spaces.

A vector space is a nonempty set $\cV$ of vectors on which we define two operations, addition and scalar multiplication, subject to the following rules.

For all $\vec{u}, \vec{v}, \vec{w}$ in $\cV$ and all scalars $c,d$ in $\R$:

  1. $\vec{u} + \vec{v}$ is in $\cV$.
  2. $\vec{u} + \vec{v} = \vec{v} + \vec{u}$.
  3. $(\vec{u} + \vec{v}) + \vec{w} = \vec{u} + (\vec{v} + \vec{w})$.
  4. There exists $\vec{0}$ (called the zero vector) in $\cV$ such that $\vec{u} + \vec{0} = \vec{u}$.
  5. For each $\vec{u}$ in $\cV$, there exists $-\vec{u}$ in $\cV$ such that $\vec{u} + (-\vec{u}) = \vec{0}$.
  6. $c \vec{u}$ is in $\cV$.
  7. $c (\vec{u} + \vec{v}) = c \vec{u} + c \vec{v}$.
  8. $(c+d) \vec{u} = c \vec{u} + d \vec{u}$.
  9. $(cd) \vec{u} = c (d \vec{u})$.
  10. $1 \vec{u} = \vec{u}$.

One can easily show the following

  • The zero vector $\vec{0}$ is unique.
  • For each $\vec{u}$, the vector $-\vec{u}$ in statement 5 is unique.
  • $0 \vec{u} = \vec{0}$ for any $\vec{u}$.
  • $-\vec{u} = (-1) \vec{u}$ for any $\vec{u}$.

Before looking at examples, note that $\R^n$ with vector addition and scalar multiplication (as defined in Section 1.3) is a vector space.

However: “Vectors” in a vector space can be loads of different things, like polynomials, continuous functions, matrices, sequences of numbers, operators, ...

Let $\cM_{2\times 2} = \cbr{\mat{cc} a & b \\ c & d \rix \:\mid\: a,b,c,d \in \R}$ be the collection of $2\times 2$ matrices with real entries. Is $\cM_{2\times 2} $ a vector space?
For $n\ge 0$, define the set of polynomials of degree at most $n$ (in the variable $t$) by \[ \cP_n \eq \cbr{p(t) = a_0 + a_1t + \dots + a_nt^n \:\mid\: a_0, \dots , a_n \in \R } .\] Is $\cP_n$ a vector space?

If $p(t) = a_0 + a_1t + \dots + a_n t^n$ and $q(t) = b_0 + b_1t + \dots + b_n t^n$, then ...

Let $\cS$ be the set of points inside and on the unit circle on the $xy$-plane: \[ \cS \eq \cbr{(x,y) \:\mid\: x^2+y^2 \le 1, \; x,y \in \R } .\] Is $\cS$ a vector space?
The set of all real-valued functions defined on $[a,b]$ is denoted \[ \cR[a,b] \eq \cbr{f \:\mid\: f: [a,b] \to \R} .\] Is $\cR[a,b]$ a vector space?

Notice that a subset of these functions are those which are continuous on $[a,b]$: \[ C[a,b] \eq \cbr{f \in \cR[a,b] \:\mid\: f \text{ is continuous} } .\]

Like in the last example, vector spaces can be formed from subsets of other vector spaces.
A set $A$ is a subset of another set $B$, denoted $A \subseteq B$, if every element in $A$ is also in $B$.

A subset of a vector space is not always another vector space. More conditions are needed.

Let $\cV$ be a vector space. A subspace $\cH$ of $\cV$ is a subset of $\cV$ such that:
  1. The zero vector $\vec{0}$ of $\cV$ is in $\cH$.
  2. For each $\vec{u}$ and $\vec{v}$ in $\cH$, then $\vec{u} + \vec{v}$ is in $\cH$. ($\cH$ is closed under addition.)
  3. For each $\vec{u}$ in $\cH$ and scalar $c \in \R $, then $c \vec{u}$ is in $\cH$. ($\cH$ is closed under scalar multiplication.)

Note that $\cH$ is a vector space on its own.

  1. Every vector space $\cV$ is a subspace of itself, and $\cbr{\vec{0}}$ is a subspace of every $\cV$.
  2. Recall the following sets:

    \begin{align*} \cR[a,b] & \eq \cbr{f \:\mid\: f: [a,b] \to \R} \\ C[a,b] & \eq \cbr{f \in \cR[a,b] \:\mid\: f \text{ is continuous} } \\ \cP_n & \eq \cbr{p(t) = a_0 + \dots + a_n t^n \:\mid\: a_0, \dots , a_n \in \R }. \end{align*}

    Then $C( \R )$ and $\cP_n$ are subspaces of $\cR(\R )$, since they contain the zero function $0(t)=0$, and the sum of two continuous functions (polynomials) and scalar multiples of a continuous function (polynomial) is also a continuous function (polynomial). Further, $\cP_n$ is a subspace of $C(\R )$.
Is the set $\cH = \cbr{\mat{c} a \\ 0 \\ b \rix \:\mid\: a, b \in \R }$ a subspace of $\R^3$?
  • The zero vector $\vec{0} = \mat{c} 0 \\ 0 \\ 0 \rix$ is in $\cH$.
  • The sum of any two vectors in $\cH$ remains in $\cH$: \[ \vec{u} + \vec{v} \eq .\]
  • For any scalar $c \in \R $ and $\vec{u} \in \cH$, \[ c \vec{u} \eq .\]
Let $\cS = \cbr{\mat{c} x \\ x +1 \rix \:\mid\: x\in \R}$. Is $\cS$ a subspace of $\R^2$?

Even though the last example was negative, it does remind us of the span of a set of vectors (in this case, only one). Under what conditions can the span of some vectors be a subspace?

Let $\vec{v}_1$ and $\vec{v}_2$ be in a vector space $\cV$, and let $\cH = \Span \cbr{\vec{v}_1, \vec{v}_2}$. Is $\cH$ a subspace of $\cV$?

If $\vec{v}_1, \dots , \vec{v}_p$ are in a vector space $\cV$, then $\Span \cbr{\vec{v}_1, \dots , \vec{v}_p}$ is a subspace of $\cV$.
Follows the same argument as the above.

We call $\Span \cbr{\vec{v}_1, \dots , \vec{v}_p}$ the subspace generated by or spanned by the vectors $\vec{v}_1, \dots , \vec{v}_p$.

Notice that our last example fails because the set was \[ \cS \eq \mat{c} 0 \\ 1 \rix + \Span \cbr{\mat{c} 1 \\ 1 \rix} \] which is not the span of $(1,1)$.

Let $\cV = \cbr{\mat{c} a + 2b \\ 2a-3b \rix \:\mid\: a,b \in \R}$. Is $\cV$ a subspace of $\R^2$?
Let $\cV = \cbr{ \mat{c} a + 2b \\ a + 1 \\ a \rix \:\mid\: a, b \in \R}$. Is $\cV$ a subspace of $\R^3$?

Recall that $\cM_{2\times 2} $ is the vector space of real-valued $2\times 2$ matrices.

Is $\cH = \cbr{\mat{cc} 2a & b \\ 3a+b & 3b \rix \:\mid\: a,b \in \R }$ a subspace of $\cM_{2\times 2} $?

Any matrix in $\cH$ can be written as \[ \mat{cc} 2a & b \\ 3a+b & 3b \rix \eq \mat{cc} 2a & 0 \\ 3a & 0 \rix + \mat{cc} 0 & b \\ b & 3b \rix \eq a \mat{rr} 2 & 0 \\ 3 & 0 \rix + b \mat{cc} 0 & 1 \\ 1 & 3 \rix .\]

Therefore, we have \[ \cH \eq \Span \cbr{ \mat{rr} 2 & 0 \\ 3 & 0 \rix, \mat{cc} 0 & 1 \\ 1 & 3 \rix } \] which makes $\cH$ a subspace of $\cM_{2\times 2} $.

Recall that $C[a,b] = \cbr{f: [a,b] \to \R \:\mid\: f \text{ is continuous} } $.

Determine if $\cH = \cbr{f \in C[a,b] \:\mid\: f(a) = f(b)} $ is a subspace of $C[a,b]$.

  • The zero function is clearly in $\cH$.
  • Let $f$ and $g$ be in $\cH$.
  • Let $c \in \R $. Then

Section 4.2: Null, Column, Row Spaces

In this section we investigate three important subspaces associated to a matrix.

The null space of an $m\times n$ matrix $A$, denoted $\Null (A)$, is the set of all solutions to $A \vec{x} = \vec{0}$. That is, \[ \Null (A) \eq \cbr{\vec{x} \in \R^n \:\mid\: A \vec{x} = \vec{0}} .\] Note that $A \vec{x} = \vec{0}$ is in $\R^m $, whereas $\vec{x} \in \R^n$.
Clearly, $\Null (A)$ contains at least $\vec{0}$. For example, given $A = \mat{rrr} 1 & -3 & -2 \\ -5 & 9 & 1 \rix$, then \[ \mat{r} 5 \\ 3 \\ -2 \rix \in \Null (A) \qquad\text{and} \qquad \mat{r} 0 \\ 0 \\ 0 \rix \in \Null (A) .\]
The null space of an $m\times n$ matrix $A$ is a subspace of $\R^n $.
  • Clearly, $\vec{0}$ is in $\Null (A)$.
  • Let $\vec{u}, \vec{v}$ be in $\Null (A)$. Then $A \vec{u} = \vec{0}$ and $A \vec{v} = \vec{0}$, so \[ A (\vec{u} + \vec{v}) \eq A \vec{u} + A \vec{v} \eq \vec{0} .\] So $\vec{u} + \vec{v} \in \Null (A)$.
  • Let $\vec{u}$ be in $\Null (A)$ and $c$ be any scalar. Then \[ A (c \vec{u}) \eq c \cdot A \vec{u} \eq c \cdot \vec{0} \eq \vec{0} .\] So $c \vec{u} \in \Null (A)$.
Therefore, $\Null (A)$ is a subspace of $\R^n $.

Find a spanning set for the null space of $A = \mat{rrrrr} -3 & 6 & -1 & 1 & -7 \\ 1 & -2 & 2 & 3 & -1 \\ 2 & -4 & 5 & 8 & -4 \rix$

Find the solution of $A \vec{x} = \vec{0}$ in terms of free variables by row reducing $\mat{c|c} A & \vec{0} \rix$: \[ \mat{rrrrr|c} -3 & 6 & -1 & 1 & -7 & 0 \\ 1 & -2 & 2 & 3 & -1 & 0\\ 2 & -4 & 5 & 8 & -4 & 0 \rix \quad \xrightarrow{\text{\normalsize Row Ops.}} \quad \mat{rrrrr|c} 1 & -2 & 0 & -1 & 3 & 0 \\ 0 & 0 & 1 & 2 & -2 & 0 \\ 0 & 0 & 0 & 0 & 0 & 0 \rix \]

Since $x_2, x_4, x_5$ are free variables, the solution is \begin{align*} \mat{c} x_1 \\ x_2 \\ x_3 \\ x_4 \\ x_5 \rix & \eq \\ & \eq \\ & \eq \end{align*}

Notice the following observations:

  1. The spanning set $\cbr{\vec{u}, \vec{v}, \vec{w}}$ in the last example is a linearly independent set. \[ x_2 \mat{c} 2 \\ 1 \\ 0 \\ 0 \\ 0 \rix + x_4 \mat{r} 1 \\ 0 \\ -2 \\ 1 \\ 0 \rix + x_5 \mat{r} -3 \\ 0 \\ 2 \\ 0 \\ 1 \rix = \mat{c} 0 \\ 0 \\ 0 \\ 0 \\ 0 \rix \qquad \implies \qquad x_2 = x_4 = x_5 = 0 .\]
  2. When $\Null (A)$ is nontrivial (i.e., contains at least one nonzero vector), then the number of vectors in the spanning set for $\Null (A)$ is the number of free variables in the equation $A \vec{x} = \vec{0}$. When $\Null (A) = \cbr{\vec{0}}$, what can be said?
Summary of Nullspace
For any $m\times n$ matrix $A$, then:
  1. $\Null (A)$ is a subspace of $\R^n $.
  2. $\Null (A)$ is the set of all vectors $x \in \R^n $ such that $A \vec{x} = \vec{0} \in \R^m$.
  3. To find an explicit description of $\Null (A)$ (i.e., a spanning set), perform row operations on $\mat{c|c} A & \vec{0} \rix$.
  4. $\Null (A) = \cbr{\vec{0}}$ if and only if $A \vec{x} = \vec{0}$ has only the trivial solution.
The column space of an $m\times n$ matrix $A$, denoted $\Col (A)$, is the set of linear combinations of the columns of $A$. That is, for $A = \mat{ccc} \vec{a}_1 & \dots & \vec{a}_n \rix$, then \[ \Col (A) \eq \Span \cbr{\vec{a}_1, \dots , \vec{a}_n} .\]

A vector in $\Col (A)$ can be written as $A \vec{x}$ for some $\vec{x}$ in $\R^n$ (Why?). Therefore, \[ \Col (A) \eq \cbr{\vec{b} \in \R^m \:\mid\: \vec{b} = A \vec{x} \quad \text{for some } x \in \R^n} \] is an alternative way to describe the column space.

The column space of an $m\times n$ matrix $A$ is a subspace of $\R^m $.
Why? Due to .

Find $A$ so that $\Col (A) = \cW = \cbr{\mat{c} x-2y \\ 3y \\ x+y \rix \:\mid\: x,y \in \R }$.

Notice that \[ \mat{c} x-2y \\ 3y \\ x + y \rix \eq .\]

Therefore, $\cW$ is given by $\Col (A)$ for \[ A \eq .\]

Observation: The range of the linear operator $T(\vec{x}) = A \vec{x}$ from $\R^2 \to \R^3$ is $\cW$. So, we also call the column space the range of the matrix.

Recall from Section 1.4 that the columns of an $m\times n$ matrix $A$ span $\R^m$ if and only if $A \vec{x} = \vec{b}$ has a solution for every $\vec{b} \in \R^m $. Hence, the following result.

The column space of an $m\times n$ matrix $A$ is all of $\R^m $ if and only if $A \vec{x} = \vec{b}$ has a solution for all $\vec{b} \in \R^m $.

Other equivalent conditions to $\Col (A) = \R^m $:

  • $A$ has a pivot position in every row.
  • Every $\vec{b} \in \R^m $ is a linear combination of the columns of $A$.
For any $m\times n$ matrix $A$:
Null Space:
  1. $\Null (A)$ is a subspace of $\R^n $.
  2. $\Null (A)$ is the set of all $x \in \R^n $ such that $A \vec{x} = \vec{0} \in \R^m$.
  3. To find a spanning set, perform row operations on $\mat{c|c} A & \vec{0} \rix$.
  4. $\Null (A) = \cbr{\vec{0}}$ if and only if $A \vec{x} = \vec{0}$ has only the trivial solution.
Column Space:
  1. $\Col (A)$ is a subspace of $\R^m $.
  2. $\Col (A)$ is the set of all $\vec{b} \in \R^m $ such that $\vec{b} = A \vec{x}$ for some $\vec{x} \in \R^n $.
  3. To find a spanning set, list the columns of $A$.
  4. $\Col (A) = \R^m $ if and only if $A \vec{x} = \vec{b}$ has a solution for all $\vec{b} \in \R^m $.
The row space of an $m\times n$ matrix $A$, denoted $\Row (A)$, is the set of linear combinations of the rows of $A$.

\[ \text{If } A = \mat{c} \vec{r}_1^T \\ \vdots \\ \vec{r}_m^T \rix, \qquad \Row(A) \eq \Span \cbr{\vec{r}_1, \dots , \vec{r}_n} .\]

For any $m\times n$ matrix $A$, $\Row(A)$ is a subspace of $\R^n $.
If $A = \mat{rrrr} -1 & 2 & 3 & 6 \\ 2 & -5 & -6 & -12 \\ 1 & -3 & -3 & -6 \rix$, then

\begin{align*} \Row(A) & \eq \Span \cbr{\mat{r} -1 \\ 2 \\ 3 \\ 6 \rix, \mat{r} 2 \\ -5 \\ -6 \\ -12 \rix, \mat{r} 1 \\ -3 \\ -3 \\ -6 \rix } \quad & \subseteq \R^4 \\ \Col (A) & \eq \Span \cbr{\mat{r} -1 \\ 2 \\ 1 \rix, \mat{r} 2 \\ -5 \\ -3 \rix, \mat{r} 3 \\ -6 \\ -3 \rix, \mat{r} 6 \\ -12 \\ -6 \rix } \quad & \subseteq \R^3 \\ A^T & \eq \mat{rrr} -1 & 2 & 1 \\ 2 & -5 & -3 \\ 3 & -6 & -3 \\ 6 & -12 & -6 \rix \\ \Col(A^T) & \eq \Span \cbr{\mat{r} -1 \\ 2 \\ 3 \\ 6 \rix, \mat{r} 2 \\ -5 \\ -6 \\ -12 \rix, \mat{r} 1 \\ -3 \\ -3 \\ -6 \rix } \quad & \subseteq \R^4 \end{align*} This reveals that:

Let $A = \mat{rrr} 1 & 2 & 3 \\ 2 & 4 & 7 \\ 3 & 6 & 10 \\ 0 & 0 & 1 \rix$, with rref of $\mat{rrr} 1 & 2 & 0 \\ 0 & 0 & 1 \\ 0 & 0 & 0 \\ 0 & 0 & 0 \rix$.
  1. $\Col (A)$ is a subspace of what?
  2. $\Row (A)$ is a subspace of what?
  3. $\Null (A)$ is a subspace of what?
  4. Find a nonzero vector in $\Col (A)$.
  5. Find a nonzero vector in $\Row(A)$.
  6. Find a nonzero vector in $\Null (A)$.
Determine if the following sets are vector spaces. Verify your answer. Hint: Use the theorems of this section!
  1. $\cV = \cbr{\mat{c} x \\ y \\ z \rix \:\mid\: x-y=0, \; y+z=0}$
  2. $\cM = \cbr{\mat{c} c-6d \\ d \\ c \rix \:\mid\: c,d \in \R } $

Section 4.3: Linearly Independent Sets, Bases

Here we answer: which subsets of vectors span a vector space as “efficiently” as possible?

Recall that a set of vectors $\cbr{\vec{v}_1, \dots , \vec{v}_p}$ in a vector space $\cV$ is linearly independent if \[ c_1 \vec{v}_1 + \dots + c_p \vec{v}_p \eq \vec{0} \] has only the trivial solution $c_1 = 0, \dots, c_p=0$.

The following theorem from Section 1.7 applies to a general vector space.

A set of two or more vectors $\cbr{\vec{v}_1, \dots , \vec{v}_p}$, with $\vec{v}_1 \neq \vec{0}$, is linearly dependent if and only if some $\vec{v}_j$ ($j>1$) is a linear combination of the preceding vectors $\vec{v}_1, \dots , \vec{v}_{j-1}$.
Are the following sets linearly independent or linearly dependent?
  1. $\cbr{p_1, p_2, p_3}$ in $\cP_2$ where $p_1(t)=t, p_2(t)=t^2, p_3(t)=4t+2t^2$.
  2. $\cbr{p_1, p_2, p_3}$ in $\cP_3$, where $p_1(t) = (t-1)$, $p_2(t) = (t-1)(t-2)$, and $p_3(t) = (t-1)(t-2)(t-3)$. Notice that \begin{align*} c_1 p_1 + c_2 p_2 + c_3 p_3 & \eq 0 \\ c_1 (t-1) + c_2 (t-1)(t-2) + c_3 (t-1)(t-2)(t-3) & \eq 0 \\ -a+2b-6c + t(a-3b+11c) + t^2(b-6c) + t^3(c) & \eq 0 \end{align*}

In Homework 2, you considered the matrix \[ A \eq \mat{rrrrr} 8 & 11 & -6 & -7 & 13 \\ -7 & -8 & 5 & 6 & -9 \\ 11 & 7 & -7 & -9 & -6 \\ -3 & 4 & 1 & 8 & 7 \rix \qquad \xrightarrow{\RREF} \quad \mat{ccccc} 1 & 0 & -7 /13 & 0 & 0 \\ 0 & 1 & -2 /13 & 0 & 0 \\ 0 & 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 0 & 1 \rix .\]

The columns of $A$ span $\R^4$ (pivot position in every row) but are not linearly independent. Indeed, $\vec{a}_3 = \textcolor{red}{-\frac{7}{13}\vec{a}_1 - \frac{2}{13} \vec{a}_2}$. This tells us that \begin{align*} \Span \cbr{\vec{a}_1, \vec{a}_2, \vec{a}_3, \vec{a}_4, \vec{a}_5} & \eq \Span \cbr{\vec{a}_1, \vec{a}_2, \textcolor{red}{-\frac{7}{13}\vec{a}_1 - \frac{2}{13} \vec{a}_2}, \vec{a}_4, \vec{a}_5} \\ & \eq \Span \cbr{\vec{a}_1, \vec{a}_2, \vec{a}_4, \vec{a}_5 }. \end{align*} Why? Let's prove a simpler version (c.f. homework 2!).

If $\cH = \cbr{\vec{v}_1, \vec{v}_2, \vec{v}_3}$ and $\vec{v}_3 = \vec{v}_1 + \vec{v}_2$, then $\Span \cH = \Span \cbr{\vec{v}_1, \vec{v}_2}$.
The goal: show 1) all vectors in $\Span \cH$ are also in $\Span \cbr{\vec{v}_1, \vec{v}_2}$, and 2) all vectors in $\Span \cbr{\vec{v}_1, \vec{v}_2}$ are in $\Span \cH$. I.e., $\Span \cH \subseteq \Span \cbr{\vec{v}_1, \vec{v}_2}$ and $\Span \cbr{\vec{v}_1, \vec{v}_2} \subseteq \Span \cH$.

First, let $\vec{x}$ be in $\Span \cH$. Then for some $c_1, c_2, c_3$, \[ \vec{x} \eq c_1 \vec{v}_1 + c_2 \vec{v}_2 + c_3 \pbr{\vec{v}_1 + \vec{v}_2} \eq (c_1 + c_3) \vec{v}_1 + (c_2 + c_3) \vec{v}_2 .\] Therefore, $\vec{x}$ is in $\Span \cbr{\vec{v}_1, \vec{v}_2}$.

On the other hand, let $\vec{y}$ be in $\Span \cbr{\vec{v}_1, \vec{v}_2}$. Then for some $c_1, c_2$, \[ \vec{y} \eq c_1 \vec{v}_1 + c_2 \vec{v}_2 \eq c_1 \vec{v}_1 + c_2 \vec{v}_2 + 0\cdot \vec{v}_3 .\] Therefore, $\vec{y}$ is in $\Span \cH$. So $\cH$ and $\cbr{\vec{v}_1, \vec{v}_2}$ span the same space.

So columns 1, 2, 4, and 5 of $A$ span $\R^4$. This is a “more efficient” spanning set for $\R^4$ compared to all the columns of $A$.

This gives us the notion of a basis set—an “efficient” spanning set in that it does not contain unnecessary vectors.

Let $\cH$ be a subspace of a vector space $\cV$. A set of vectors $\cB$ in $\cV$ is a basis for $\cH$ if
  1. $\cB$ is a linearly independent set, and
  2. $\cH = \Span (\cB)$.
That is: a basis is a linearly independent spanning set for the subspace.
A basis for $\R^4$ is given by \[ \cB \eq \cbr{\vec{e}_1, \vec{e}_2, \vec{e}_3, \vec{e}_4} \] which is called the standard basis for $\R^4$. Why? Because \[ A \eq \mat{cccc} \vec{e}_1 & \vec{e}_2 & \vec{e}_3 & \vec{e}_4 \rix \eq \mat{cccc} 1 & 0 & 0 & 0 \\ 0 & 1 & 0 & 0 \\ 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 1 \rix \] has a pivot in every row, so its columns span $\R^4$, and there is a pivot in every column, so its columns are linearly independent.
Show that $\cB = \cbr{1, t, t^2, \dots , t^{n}}$ is a basis for $\cP_n$.

Clearly, $\cB$ spans $\cP_n$ because any polynomial in $\cP_n$ can be written as

for some choice of $c_0, \dots , c_n$ coefficients. And, $\cB$ is linearly independent because

implies that all coefficients $c_0, \dots , c_n$ are $0$ (by simply matching).

The following (very important!) theorem generalizes .

Let $\cS = \cbr{\vec{v}_1, \dots , \vec{v}_p}$ be a set of vectors in a vector space $\cV$ and $\cH = \Span \cbr{\vec{v}_1, \dots , \vec{v}_p}$.
  1. Suppose one of the vectors in $\cS$, say $\vec{v}_k$, is a linear combination of the other vectors in $\cS$. Then $\cS$ without this vector $\vec{v}_k$ still spans $\cH$.
  2. If $\cH \neq \cbr{\vec{0}}$, then some subset of $\cS$ is a basis for $\cH$.

That is, by removing a vector that is a linear combination of the others, you obtain a new set which still spans the same space.

This theorem tells us that we can construct a basis for a vector space $\cV$ by starting with a spanning set of $\cV$ and then pruning it down to a linearly independent set.

In this view, a basis is a spanning set that is as small as possible. If you keep removing vectors from a spanning set after you have removed enough to make it linearly independent, then the vector you delete would not be a linear combination of the smaller set. Hence, you no longer have a spanning set and thus no longer a basis.

Alternatively, a basis is a linearly independent set that is as large as possible. If $\cS$ is linearly independent and spans $\cV$, and then you add another vector from $\cV$ into $\cS$, the enlarged set becomes linearly dependent and therefore not a basis.

Given a matrix, what is a basis for its nullspace?

For illustration, recall : \[ A=\mat{rrrrr|c} -3 & 6 & -1 & 1 & -7 & 0 \\ 1 & -2 & 2 & 3 & -1 & 0\\ 2 & -4 & 5 & 8 & -4 & 0 \rix \quad \xrightarrow{\text{\normalsize Row Ops.}} \quad \RREF(A)= \mat{rrrrr|c} 1 & -2 & 0 & -1 & 3 & 0 \\ 0 & 0 & 1 & 2 & -2 & 0 \\ 0 & 0 & 0 & 0 & 0 & 0 \rix \]

Since $x_2, x_4, x_5$ are free variables, the solution is \begin{align*} \mat{c} x_1 \\ x_2 \\ x_3 \\ x_4 \\ x_5 \rix & \eq x_2 \mat{c} 2 \\ 1 \\ 0 \\ 0 \\ 0 \rix + x_4 \mat{r} 1 \\ 0 \\ -2 \\ 1 \\ 0 \rix + x_5 \mat{r} -3 \\ 0 \\ 2 \\ 0 \\ 1 \rix \eq x_2 \vec{u} + x_4 \vec{v} + x_5 \vec{w} . \end{align*} Therefore, we have \[ \Null (A) \eq \]

What is a basis for the column space? Observe that \begin{align*} \vec{a}_2 & = -2 \vec{a}_1 & \quad \vec{b}_2 &= -2 \vec{b}_1 \\ \vec{a}_4 & = -\vec{a}_1 + 2 \vec{a}_3 & \quad \vec{b}_4 & = - \vec{b}_1 + 2 \vec{b}_3 \\ \vec{a}_5 & = 3 \vec{a}_1 -2 \vec{a}_3 & \quad \vec{b}_5 & = 3 \vec{b}_1 -2 \vec{b}_3 .\end{align*} Elementary row operations do not affect linear dependence relations among the columns!

Since a spanning set for $\Col(A)$ is all the columns of $A$, but only columns 1 and 3 are linearly independent, then by the Spanning Set Theorem \[ \Span \cbr{\vec{a}_1, \vec{a}_2, \vec{a}_3, \vec{a}_4, \vec{a}_5} \eq \]

In summary, we have the following.

The pivot columns of a matrix form a basis for its column space.
Warning! The pivot columns of $A$ form the basis, not the columns in the RREF.

What about a basis for the row space?

Two row equivalent matrices $A$ and $B$ have the same row space. Furthermore, if $B = \RREF(A)$, then the nonzero rows of $B$ form a basis for $\Row(A)$.

Continuing from the matrix in : \[ A=\mat{rrrrr} -3 & 6 & -1 & 1 & -7 \\ 1 & -2 & 2 & 3 & -1 \\ 2 & -4 & 5 & 8 & -4 \rix \quad \xrightarrow{\text{\normalsize Row Ops.}} \quad B= \mat{rrrrr} 1 & -2 & 0 & -1 & 3 \\ 0 & 0 & 1 & 2 & -2 \\ 0 & 0 & 0 & 0 & 0 \rix .\] A basis for the row space of $A$ is:

Summary: Given $A$ and its $\RREF$, then:
  1. A basis for $\Null (A)$ consists of the vectors used in the parametric form of the solution to $A \vec{x} = \vec{0}$.
  2. A basis for $\Col (A)$ consists of the pivot columns of $A$.
  3. A basis for $\Row(A)$ consists of the transposes of the nonzero rows in the $\RREF$.
The matrix $A$ and its $\RREF$ are given below: \[ A = \mat{rrr} 1 & -2 & 3 \\ 2 & -4 & 6 \\ -3 & 6 & 9 \rix \qquad \RREF(A)= \mat{rrr} 1 & -2 & 0 \\ 0 & 0 & 1 \\ 0 & 0 & 0 \rix .\] Find bases for $\Null (A)$, $\Col (A)$, $\Row(A)$.

Solving $A \vec{x} = \vec{0}$ yields \[ \mat{c} x_1 \\ x_2 \\ x_3 \rix \eq \] Therefore, a basis for $\Null (A)$ is

A basis for $\Col (A)$ is

A basis for $\Row(A)$ is

Section 4.5: Dimension of a Vector Space

How “large” is a vector space? What is a good measure to quantify its size?

We will answer this after two insightful results.

If a vector space $\cV$ has a basis of $n$ vectors, then any set in $\cV$ containing more than $n$ vectors must be linearly dependent.

Why? Think about $\R^n $. Clearly a set of $n+1$ vectors from $\R^n $ must be linearly dependent! A proof for a general vector space requires more tools than we currently have.

Put another way: If a vector space $\cV$ has a basis of $n$ vectors, then every linearly independent subset of $\cV$ has no more than $n$ vectors.

Be Careful! If $\cV$ has a basis of $n$ vectors, then a subset of $\cV$ with at most $n$ vectors may or may not be linearly dependent.
If a vector space $\cV$ has a basis of $n$ vectors, then every basis of $\cV$ consists of exactly $n$ vectors.
Suppose $\cB_1$ is a basis with $n$ vectors and $\cB_2$ is any other basis. Because $\cB_2$ is linearly independent, then by we know $\cB_2$ has no more than $n$ vectors.
  • If $\cB_2$ has exactly $n$ vectors, we are done.
  • If $\cB_2$ has less than $n$ vectors, then applying to $\cB_2$ implies that $\cB_1$ must be linearly dependent (since $\cB_1$ has more vectors than $\cB_2$), which is a contradiction.

Therefore, $\cB_1$ and $\cB_2$ have the same number of vectors $n$.

This result means that the number of elements in any basis of a vector space is always the same. Even though a vector space can have many bases, they must all have the same number of vectors!

This allows us to quantify the “size” of the vector space.

Suppose $\cV$ is a vector space spanned by a finite set. Then we say $\cV$ is finite dimensional, and the dimension of $\cV$, written $\Dim (\cV)$, is the number of vectors in any basis for $\cV$.
  • The dimension of $\cV = \cbr{\vec{0}}$ is defined to be $0$.
  • If $\cV$ is not spanned by a finite set, we say $\cV$ is infinite dimensional.
Determine the dimensions of the following vector spaces and provide a basis.
  • $\R^2$
  • $\R^n $
  • $\cP_2$
  • $\cP_n$
  • $\cM_{2\times 2} $
  • $\cP$ (polynomials of all degrees)
Find a basis and the dimension of the subspace \[ \cW \eq \cbr{\mat{c} a + b + 2c \\ 2a + 2b + 4c + d \\ b + c + d \\ 3a + 3c + d \rix \:\mid\: a,b,c,d \in \R } .\] Notice that we can write \[ \mat{c} a + b + 2c \\ 2a + 2b + 4c + d \\ b + c + d \\ 3a + 3c + d \rix \eq \]
Let $\cH$ be a subspace of a finite dimensional vector space $\cV$. Any linearly independent subset in $\cH$ can be expanded, if necessary, to a basis for $\cH$. Furthermore, $\cH$ is finite dimensional and \[ \Dim (\cH) \eq[\le] \Dim (\cV) .\]
Read the textbook (page 243).
Start Action Result
Spanning set Remove L.D. vectors Create a basis
Spanning set Add vectors Cannot create a basis
L.I. set Remove vectors Cannot create a basis
L.I. set Add vectors Create a basis
Let $\cH = \Span \cbr{\vec{v}_1, \vec{v}_2}$ be a subspace of $\R^3$, where $\vec{v}_1 = \mat{c} 1 \\ 0 \\ 0 \rix, \; \vec{v}_2 = \mat{c} 1 \\ 1 \\ 0 \rix$.
  1. What is the dimension of $\cH$?
  2. Is $\cH$ a basis for $\R^3$? If not, create one using $\vec{v}_1 $ and $\vec{v}_2$.

If you know the dimension of a vector space (say, $p$), you can find a basis by:

  1. Finding a set of $p$ vectors, and
  2. Checking if the set is linearly independent OR checking if the set spans the space.

Namely, if you have the right number of vectors you do NOT have to check both!

Let $\cV$ be a $p$-dimensional vector space, $p\ge 1$. Then:
  1. Any set of $p$ linearly independent vectors is a basis for $\cV$.
  2. Any spanning set of $p$ vectors is a basis for $\cV$.
Read the textbook (page 243).
Show that a basis for $\cP_2$ is $\cbr{t, \, 1-t, \, 1+t-t^2}$.

We know $\cP_2$ has dimension $3$, which is the number of elements in this set. Therefore, we can check (via The Basis Theorem) if the set is either 1) linearly independent, or 2) a spanning set.

For linear independence, form the homogeneous equation \begin{align*} c_1 \cdot t + c_2 \cdot (1-t) + c_3 \cdot (1+t-t^2) & \eq 0 \\ (c_2 + c_3) \cdot 1 + (c_1-c_2+c_3)\cdot t + (-c_3) \cdot t^2 &\eq 0 .\end{align*} Matching coefficients, then the only solution is $c_3=0$ (from $t^2$) and so $c_2=0$ (from $1$). Therefore, $c_2=0$ and the set is linearly independent; so it is a basis.

Showing that the set spans $\cP_2$ is very difficult!

Recall how we found bases for $\Null (A), \Col (A), \Row(A)$:

Given $A$ and its $\RREF$, then:
  1. A basis for $\Null (A)$ consists of the vectors used in the parametric form of the solution to $A \vec{x} = \vec{0}$.
  2. A basis for $\Col (A)$ consists of the pivot columns of $A$.
  3. A basis for $\Row(A)$ consists of the transposes of the nonzero rows in $\RREF(A)$.

The dimensions of these spaces are very important!

The rank of an $m\times n$ matrix is the dimension of its column space. The nullity of an $m\times n$ matrix is the dimension of its null space.
  • A basis for $\Col (A)$ are the pivot columns of $A$. Therefore, \[ \Rank (A) \eq \Dim \Col(A) \eq \]
  • A basis for $\Row(A)$ are the pivot rows (transposed) from the $\RREF$ of $A$. The number of pivot rows is equal to the number of pivot columns, so \[ \Dim \Row (A) \eq \Dim \Col (A^T) \eq \] (Recall: $\Col (A^T) = \Row(A)$.)
  • A basis for $\Null (A)$ are the vectors used in the parametric representation for the solution to $A \vec{x} = \vec{0}$, so \[ \mathrm{Nullity}(A) \eq \Dim \Null (A) \eq \]
Because a column of $A$ is either a pivot column or not, we have the important result.
If $A$ is an $m\times n$ matrix, then \[ \Rank (A) + \mathrm{Nullity}(A) \eq \text{ number of columns of } A \eq n .\]
Find rank and nullity of $A = \mat{rrrrr} -3 & 6 & -1 & 1 & -7 \\ 1 & -2 & 2 & 3 & -1 \\ 2 & -4 & 5 & 8 & -4 \rix$, with $\RREF$ $\mat{rrrrr} 1 & -2 & 2 & 3 & -1 \\ 0 & 0 & 1 & 2 & -2 \\ 0 & 0 & 0 & 0 & 0 \rix$.
Suppose $A$ is $5\times 8$ with rank $5$. Find $\mathrm{Nullity}(A)$, $\Dim \Row(A)$, $\Rank (A)$, and $\Rank (A^T)$.
Suppose $A$ is a $9\times 12$ matrix. What is the smallest possible nullity of $A$?
A homogeneous system of $50$ equations in $54$ variables has $4$ free variables. Does every nonhomogeneous system have a solution?

With our knowledge of rank, column space, and null space, we can extend the Invertible Matrix Theorem from Section 2.3.

Let $A$ be an $n\times n$ matrix. Then the following are equivalent (i.e., either all statements are true or all are false).
  1. $A$ is an invertible matrix.
  2. $A$ is row equivalent to the $n\times n$ identity matrix $I_n$.
  3. $A$ has $n$ pivot positions.
  4. The equation $A \vec{x} = \vec{0}$ has only the trivial solution.
  5. The columns of $A$ form a linearly independent set.
  6. The equation $A \vec{x} = \vec{b}$ has a unique solution for all $\vec{b} \in \R^n $.
  7. The columns of $A$ span $\R^n $.
  8. There exists an $n\times n$ matrix $C$ such that $CA = I_n$.
  9. There exists an $n\times n$ matrix $D$ such that $AD = I_n$.
  10. $A^T$ is an invertible matrix.
  11. The columns of $A$ form a basis of $\R^n $.
  12. $\Col (A) = \R^n $.
  13. $\Rank (A) = n$.
  14. $\mathrm{Nullity}(A) = 0$.
  15. $\Null (A) = \cbr{\vec{0}}$.
Warning! Just as before, the IMT applies only to square matrices.
The matrix $A = \mat{cc} 1 & 1 \\ 2 & 2 \rix$ has $\RREF$ $\mat{cc} 1 & 1 \\ 0 & 0 \rix$. Then,

Basis for $\Row(A) = \Col (A^T)$: $\cbr{\mat{c} 1 \\ 1 \rix}$

Basis for $\Col (A)$: $\cbr{\mat{c} 1 \\ 2 \rix} $

Basis for $\Null (A)$: $\cbr{\mat{r} 1 \\ -1 \rix} $

Basis for $\Null (A^T)$: $\cbr{\mat{r} 2 \\ -1 \rix}$