Linear Algebra Primer

Daniel S. Stutts, Ph.D.

Original Edition: 12/1991 Current Edition: 4/1/04

1 Introduction
This primer was written to provide a brief overview of the main concepts and methods in elemen-
tary linear algebra. It was not intended to take the place of any of the many elementary linear
algebra texts in the market. It contains relatively few examples and no exercises. The interested
reader will find more in depth coverage of these topics in introductory text books. Much of the
material including the order in which it is presented comes from Howard Anton’s “Elementary
Linear Algebra” 2nd Ed., John Wiley, 1977. Another excellent basic text is “Linear Algebra and
Its Applications,” by Charles G. Cullen. A more advanced text is “Linear Algebra and its Appli-
cations” by Gilbert Strang.

The author hopes that this primer will answer some of your questions as they arise, and provide
some motivation (prime the pump, so to speak) for you to explore the subject in more depth. At
the very least, you now have a list (albeit a short one) of references from which to obtain more in
depth explanation.

It should be noted that the examples given here have been motivated by the solution of consistent
systems of equations which have an equal number of unknowns and equations. Therefore, only the
analysis of square (n by n) matrices have been presented. Furthermore, only the properties of real
matrices (those with real elements) have been included.

1.1 Explanation of Notation Used

For clarity of notation, bold symbols are used to denote vectors and matrices. For matrices, upper
case bold letters are used, and for vectors, which are n × 1 matrices, bold lower case letters are
used. Non-bold symbols are used to denote scalar quantities.

Subscripts are used to denote elements of matricies or vectors. Superscripts (when not referring
to exponentiation) are used to identify eigenvectors and their respective components.

1.1.1 Indicial Notation

A matrix A may be described by indicial notation. The term located at the ith row and j th column
is denoted be the scalar aij .

Thus, ij th component of the sum of two matrices, A and B, may be written: [A+B]ij = aij +bij

Example 1
 
(a11 + b11) (a12 + b12)
C = A(2×2) + B(2×2) =
(a21 + b21) (a22 + b22)

Hence, for example, c12 = a12 + b12.

1
2 Linear Systems of Equations
The following systems of equations

a11x11 + a12 x12 + · · · + aln xln = b1

a21 x21 + a22 x22 + · · · + a2n x2n = b2
.. (1)
.
an1 xnl + an2 xn2 + · · · + ann xnn = bn

may be written in the matrix form as

     
a11 a12 · · · aln 
 x 1 
 
 b 1 

  x2 
 
 b2 
 a21 a22 · · · a2n 

 
 ..  .. = .. (2)
 .  
 .  
 . 


 
 
 

an1 an2 · · · ann xn bn
or
Ax = b (3)
where A is a n by n matrix and x and b are n by 1 matrices or vectors.

While the solution of systems of linear equations provides one significant motivation to study
matrices and their properties, there are numerous other applications for matrices. All applications
of matrices require a reasonable degree of understanding of matrix and vector properties.

3 Matrix Properties and Definitions

For any matrices A, B, and C, the following hold:

1. A + B = B + A

2. A + (B + C) = (A + B) + C

3. A(B + C) = AB + AC

Provided that AB and AC are defined

Definition 1 AB is defined if B has the same number of rows as A has columns.

A(m×r) B(r×n) = AB(m×n)

4. Identity Matrix: AI = IA = A where

 
1 0 0 . .
 0 1 0 . . 
 
I= 
 0 0 1 . . 

 . . . . . 
. . . . 1

2
 5. Zero Matrix: 0A = A0 = 0
 
0 0 0 . .
 0 0 0 . . 
 
0= 
 0 0 0 . . 

 . . . . . 
. . . . 0

6. A + 0 = A

3.1 The transpose operation and its properties

The transpose of a matrix A, written as AT , is the matrix A with its off-diagonal components
reflected across the main diagonal. Hence, defining the components of the matrix A as aij in
indicial notation, the components of AT are given by aji .

3.1.1 Properties of the transpose operation

1. (AT )T = A
2. (A + B)T = AT + BT
3. (kA)T = kAT
4. (AB)T = BT AT
A matrix A is termed symmetric if AT = A and skew-symmetric if AT = −A. Clearly, a skew-
symmetric matrix can only have zero diagonal terms.

3.2 Multiplication of a Matrix

There are a few rules for matrix and vector multiplication which must be considered in addition to
the rules for the more familiar scalar algebra. These are enumerated with examples below.
1. Multiplication by a scalar The product of a scalar and the matrix with each of its elements
multiplied by the scalar. For example:
 
αa11 αa12
αA = [αaij ] =
αa21 αa22
2. Multiplication by a matrix The product of two matrices, A and B, when it is defined is
another matrix, C
C = AB (4)
where the components of C may be computed as follows:
n
X
cij = aik bkj = cij (5)
k=1

Example 2
     
a11 a12 b11 b12 a11b11 + a12b21 a11b12 + a12b22
AB = =
a21 a22 b21 b22 a21b11 + a22b21 a21b12 + a22b22

3
 Note that in general, matrices are not commutative over multiplication:

AB 6= BA (6)
This fact leads to the definition of pre-multiplication and post-multiplication:

Definition 1 A matrix A is said to be pre-multiplied by a matrix B when B multiplies A

from the left, and post-multiplied when B is multiplied by A from the left.

Other terminology for the direction of multiplication in common use is left multiplication and
right multiplication. In Equation (6), on the r.h.s., A is being pre-multiplied by B, and on
the l.h.s. A is being post-multiplied by B.

3. Multiplication by a vector:

     
a11 a12 x1 a11x1 + a12 x2
Ax = =
a21 a22 x2 a21x1 + a22 x2

Pre-multiplication of a matrix by a vector requires taking the transpose of the vector first in
order to comply with the rules of matrix multiplication.

3.3 Matrix Inverse

A−1 A = I if A−1 exists,

3.4 The Determinant Operation

Recall from vector mechanics that in three space (dimensions) the vector product of any two vectors
A = (a1i + a2 j + a3k) and B = (b1i + b2j + b3k) is defined as

 
i j k
A × B = det  a1 a2 a3 
b1 b2 b3

= (a2b3 − a3 b2)i − (a1b3 − a3 b1)j + (a1 b2 − a2 b1)k (7)

This method of computing the determinant is called cofactor expansion. A cofactor is the signed
minor of a given element in a matrix. A minor Mij is the determinant of the sub matrix which
remains after the ith row and the j th column of the matrix are deleted. In this case, we have

M11 = (a2 b3 − a3b2 ) (8)

M12 = (a1 b3 − a3b1 ) (9)

4
 M13 = (a1 b2 − a2b1 ) (10)
The cofactors are given by

Cij = (−1)i+j Mij (11)

Hence,
C11 = (−1)1+1M11 = M11 (12)
C12 = (−1)1+2 M12 = −M12 (13)

etc.
In the above example,
 
i j k
det  a1 a2 a3  = c11i + c12j + c13k (14)
b1 b2 b3

 
2 1
Example 3 Let A = then
3 −2

detA = 2(−2) − 1(3) = −7

 
2 1 0
Example 4 Let A =  3 −2 1 
1 −1 2
then

detA = 2(−2(2) − (−1)(1)) − 1(3(2) − 1(1)) + 0(3(−1) − (−2)(1)) = 2(−3) − 5 = −11

Expansion by cofactors is mostly useful for small matrices (less than 4×4). For larger matrices, the
number of operations becomes prohibitively large. For example:

A 2 by 2 matrix requires 2 multiplications

A 3 by 3 matrix requires 9 multiplications

However, a 4 by 4 matrix requires the computation of 4+4! = 28 signed elementary products.

A 10 by 10 matrix would require 10 + 10! = 3,628,810 signed elementary products!

This trend suggests that soon even the largest and fastest computers would choke on such a
computation.

5
 For large matrices, the determinant is best computed using row reduction.

Row reduction consists of using elementary row and column operations to reduce a matrix down
to a simpler form, usually upper or lower triangular form.

This is accomplished by multiplying one row by a constant and adding it another row to produce
a zero at the desired position.

 
2 1 0
Example 5 Let A =  3 −2 1 
1 −1 2
Reduce A to upper triangular form, i.e., all zeros under the main diagonal (2 -2 2).

Multiplying row 1 by -1/2 and adding it to row 3 yields

 
2 1 0
 3 −2 1 
0 −1.5 2

Similarly, multiplying row 1 by -3/2 and adding it to row 2 yields

 
2 1 0
 0 −3.5 1 
0 −1.5 2
15
multiplying row 2 by − 3.5 = − 37 and adding it to row 3
 
2 1 0
 0 −7 1 
2
0 0 11 7

The determinant is now easily computed by multiplying the elements of the main diagonal.

detA = 2(− 72 )( 11
7 ) = −11

This type of row reduction is called Gaussian elimination and is much more efficient than the co-
factor expansion technique for large matrices.

3.5 Properties of Determinant Operations

1. If A is a square matrix then det(AT ) = det(A)

2. det(kA) = kn det(A) where A is an n × n matrix and k is a scalar

3. det(AB) = det(A)det(B) where A and B are square matrices of the same size.

4. A square matrix A is invertible if and only if det(A) 6= 0

Proof 1 If A is invertible then AA−1 = I ⇒ det AA−1 = (detA) detA−1 = detI = 1
1
Thus, detA−1 = detA ⇒ detA 6= 0. 

6
 An important implication of this result is the following.

5. For a homogeneous system of linear equations Ax = 0

There exists a nontrivial x (x 6= 0) if and only if det(A) = 0

Proof 2 Assume det(A) 6= 0. This implies that A−1 exists.

A−1 Ax = Ix = 0. This implies that x = 0, (a contradiction), so the only way that x can be
other than zero is for det(A) = 0, and hence, for non-trivial x, A−1 not to exist. 

This result is used very often in applied mathematics, physics and engineering.

6. If A is invertible then
1
A−1 = det(A) adj(A) where adjA is the adjoint of A.

Definition 2 The adjoint of a matrix A is defined as the transpose of the cofactor matrix of
A.

Another way to calculate the inverse of a matrix is by Gaussian elimination. This method is
easier to apply on larger matrices.

Since A−1 A = I, we start with the matrix A which we want to invert on the left and the
identity matrix on the right. We then do elementary row operations (Gaussian Elimination) on the
matrix while simultaneously doing the same operations on I. This can be accomplished by adjoining
the two matrices to form a matrix of the form [A I].

Example 6  
1 2 3
A= 2 5 3  (15)
1 0 8
Adjoining A with I yields  
1 2 3 1 0 0
 2 5 3 0 1 0  (16)
1 0 8 0 0 1
Adding -2 times the first row to the second row and -1 times the first row to the third yields
 
1 2 3 1 0 0
 0 1 −3 −2 1 0  (17)
0 −2 5 −1 0 1

Adding 2 times the 2nd row to the third yields

 
1 2 3 1 0 0
 0 1 −3 −2 1 0 
0 0 1 −5 2 1
Multiplying the third row by -1 yields

7
  
1 2 3 1 0 0
 0 1 −3 −2 1 0 
0 0 1 5 −2 −1
Adding 3 times the third row to the second and -3 times the third row to the first yields

 
1 2 3 −14 6 3
 0 1 −3 13 −5 −3 
0 0 1 5 −2 −1

Finally adding -2 times the second row to the first yields:

 
1 0 0 −40 16 9
 0 1 0 13 −5 −3 
0 0 1 5 −2 −1
Thus,
 
−40 16 9
A−1 =  13 −5 −3 
5 −2 −1

3.5.1 Cramer’s Rule

If Ax = b is a system of n linear equations in n unknowns such that det(A) 6= 0 then, the system
has a unique solution which is given by

det(A1 ) det(A2) det(A3 )

x1 = , x2 = , x3 = (18)
det(A) det(A) det(A)
where Aj is the matrix obtained by re placing the entries of the j th column A by the entries in the
vector b, where
 
b1
 b2 
 
 . 
b=


 (19)
 . 
 . 
bn
Example 7 Consider the following system:

2x1 + 4x2 − 2x3 = 18

2x2 + 3x3 = −2
x1 + 5x3 = −7
solve for x1 , x2 and x3 .

Solution: Recasting the system in matrix-vector form, we have

    
2 4 −2 x1 18
 0 2 3   x2  =  −2 
1 0 5 x3 −7

8
Next, we define the three matrices formed by replacing in turn each of the columns of A with b:

   
18 4 −2 2 18 −2
A1 =  −2 2 3  A2 =  0 −2 3 
−7 0 5 1 −7 5

 
2 4 18
A3 =  0 2 −2 
1 0 −7
Next, we compute the individual determinants:

det(A) = 2[2(5) − (−2(1)] − 3[2(0) − 4(1)] = 36

det(A1) = (−7)[4(3) − (−2)(2)] + 5[18(2) − 4(−2)] = 108
det(A2) = 2[−2(5) − (3)(−7)] + 1[18(3) − (−2)(−2)] = 72
det(A3) = 2[2(−7) − (−2)(0)] + 1[4(−2) − 18(2)] = −72

Thus,

det(A1) 108 det(A2 ) 72 det(A3) 72

x1 = = = 3, x2 = = = 2, x3 = = − = −2 (20)
det(A) 36 det(A) 36 detA) 36
Note that we took advantage of any zero elements by expanding the cofactors along the rows or
columns that contained them.

Cramer’s rule is particularly efficient to use on 2 × 2 system.

Consider a general 2nd-order example

Example 8     
a11 a12 x1 b1
= (21)
a21 a22 x2 b2
det(A) = a11a22 − a12a21 , det(A1 ) = b1a22 − a12 b2, det(A2 ) = a11 b2 − b1a21 (22)
Thus,
det(A1) a22b1 − a12b2 det(A2 ) a11 b2 − b1a21
x1 = = , x2 = = (23)
det(A) a11 a22 − a12a21 det(A) a11 a22 − a12a21

3.6 The Characteristic Polynomial and Eigenvalues and Vectors

We have shown that the system
Ax = λx (24)
where λ is a scalar value. Equivalently, we may write

(A − λI)x = 0 (25)
Hence, from Theorem (5) on page 7, there exists a nontrivial x if and only if

det(A − λI) = 0 (26)

9
 Evaluation of the above results in a polynomial in λ. This is the so called characteristic poly-
nomial and its roots λi are the characteristic values or eigenvalues. Evaluation of (26) yields,

λn + c1λn−1 + c2λn−2 · · · + cn = 0 (27)

Furthermore, the solution of (25), xi, corresponding to the ith eigenvalue, is the ith eigenvector of
the matrix A. It can be shown that the matrix thA itself satisfies the characteristic polynomial.

An + c1 An−1 + c2 An−2 + · · · + cn I = 0

This result is known as Cayley-Hamilton Theorem. It may be shown that the matrix A is also anihi-
lated by a minimum polynomial of degree less than or equal to that of the characteristic polynomial.

Example 9 Find the eigenvalues and eigenvectors of

 
4 −5
A=
1 −2
Solution:
     
4 −5 1 0 4−λ −5
det −λ = det
1 −2 0 1 1 −2 − λ

= (4 − λ)(−2 − λ) + 5 = 0

λ2 − 2λ − 3 = 0
or

(λ − 3)(λ + 1) = 0
Thus, λ1 = −1, and λ2 = 3 The eigenvector of A corresponding to λ = −1 may be found as

follows:
    
4 − (−1) −5 x11 0
(A − λI) x1 = =
1 −2 − (−1) x12 0
or

    
5 −5 x11 0
=
1 −1 x12 0
which has an obvious solution of

 
1 1
x = c1
1

where c1 is any scalar.

10
 Similarly, substituting λ2 = 3 yields

    
1 −5 x21 0
=
1 −5 x22 0

Thus,

 
2 5
x = c2
1

The calculation of eigenvalues and eigenvectors has many applications. One important application
is the similarity transformation. Under certain conditions, a general system of equations may be
transformed into a diagonal system. The most important case is that of symmetric matrices which
will be discussed later. In other words, the system Ax = b may be transformed into an equivalent
system Dy = c where D is the diagonal matrix – making the solution of Dy = c especially easy.
Two matrices A and B are said to be similar if det(A) = det(B). Another way of saying the above
is if A and B are square matrices B is similar to A if and only if there is an invertible matrix P
such that A = PBP−1 . It turns out that an n × n matrix A is diagonalizable if A has n linearly
independent eigenvectors.

Example 10 Since the previous example had two independent eigenvectors (i.e., x1 6= Kx2 for
any scalar K)
 
4 −5
A= (28)
1 −2

should be diagonalizable. A matrix composed of x1 and x2 as its two columns will diagonalize A.
We show that this is so by trying it!

   
−1 − 14 5
4 −4 5 1 5
P AP = 1
4 − 14 1 −2 1 1

    
1 −1 5 −1 15 −1 0
= =
4 1 −1 1 −3 0 3

We see that D is composed of λ1 and λ2 on its main diagonal and zeros elsewhere. Hence, the
matrix, A, was indeed similar to a diagonal matrix.

11
3.7 Special Properties of Symmetric Matrices
Symmetric matrices have several special properties. The principal ones for an n by n symmetric
matrix are enumerated below:

1. Symmetric real matrices (those with real elements) have n real eigenvalues.

2. Eigenvectors corresponding to distinct real roots are orthogonal.

Proof 3 If A = AT , and
Ax1 = λ1x1 (29)
and
Ax2 = λ2x2 (30)
Then,
λ1 x2T Ax1 = λ1x2T x1 (31)
and
λ2 x1T Ax2 = λ2x1T x2 (32)
Since x2T x1 = x1T x2 and
x2T Ax1 = x1T Ax2 (33)
subtraction of (31) from (32) yields

(λ2 − λ1) x1T x2 = 0 (34)

Hence, if λ1 6= λ2 , x1 and x2 are orthogonal as claimed. 

3. If λr is a root of the characteristic polynomial of algebraic multiplicity m, there exist m

independent eigenvectors corresponding to λr .

One of the most important consequences of the above for symmetric matrices is that all symmetric
matrices are similar to a diagonal matrix. This fact has powerful consequences in the solution of
systems of linear ordinary differential equations with constant coefficients which result from the
application of Newton’s 2nd law, or Hamilton’s principle. Essentially, such systems, which usually
result from symmetric operators, may be uncoupled by a similarity transformation, and hence, each
ordinary differential equation solved individually. Exceptions to this rule include systems modeled
with general viscous damping, and those with gyroscopic inertial terms.

The following examples were contributed by Dr. Geroid P. MacSithigh.

Example 11 (symmetric)
 
4 0 0
A= 0 4 0 
0 0 5
Characteristic polynomial: (λ − 4)2(λ − 5) = 0

λ1 = 5, λ2 = λ3 = 4

12
 λ1 = 5

 
1
x1 =  −1 
0

   
1 0
λ2 = 4, x2 =  1  , x3 =  0  (35)
0 1

Example 12 (non-symmetric)
 
5 0 0
A= 4 5 0 
0 0 3
Characteristic polynomial: (λ − 5)2(λ − 3) = 0

λ1 = 3, λ2 = λ3 = 5

 
1
λ1 = 3, x1 =  1 
0

 
0
λ2 = 5; x2 =  1 
0

Thus, only one eigenvector corresponding to λ = 5.

13
4 Bibliography
1. Anton, Howard, “Elementary Linear Algebra,” 1977, John Wiley&Sons

2. Cullen, Charles G., “Matrices and Linear Transformations,” 1972, 2nd Ed., Addison-Wesley,
Reprinted by Dover 1990.

3. Strang, Gilbert, “Linear Algebra and Its Applications,” 1988, 3rd Ed., International Thomson
Publishing.

14