You are on page 1of 111

ENG1091

Mathematics for Engineering


Lecture notes
Clayton Campus
2013 Campus
Australia Malaysia South Africa Italy India monash.edu/science
School of Mathematical Sciences Monash University
Contents
1. Vectors in 3-dimensions 3
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.2 Algebraic properties . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Vector Dot Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1 Unit Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Vector Cross Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3.1 Interpreting the cross product . . . . . . . . . . . . . . . . . . . . . 7
1.3.2 Right Hand Thumb rule . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Scalar and Vector projections . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4.1 Scalar Projections . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4.2 Vector Projection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2. Three-Dimensional Euclidean Geometry. Lines. 11
2.1 Lines in 3-dimensional space . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Vector equation of a line . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3. Three-Dimensional Euclidean Geometry. Planes. 14
3.1 Planes in 3-dimensional space . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.1.1 Constructing the equation of a plane . . . . . . . . . . . . . . . . . 15
3.1.2 Parametric equations for a plane . . . . . . . . . . . . . . . . . . . 16
3.2 Vector equation of a plane . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4. Linear systems of equations 19
4.1 Examples of Linear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.1.1 Bags of coins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.1.2 Silly puzzles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.1.3 Intersections of planes . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.2 A standard strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.3 Lines and planes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5. Gaussian Elimination 25
5.1 Gaussian elimination and back-substitution . . . . . . . . . . . . . . . . . 26
5.2 Gaussian elimination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
26-Jul-2013 2
School of Mathematical Sciences Monash University
5.2.1 Gaussian elimination strategy . . . . . . . . . . . . . . . . . . . . . 27
5.3 Exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
6. Matrices 29
6.1 Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
6.1.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
6.1.2 Operations on matrices . . . . . . . . . . . . . . . . . . . . . . . . . 31
6.1.3 Some special matrices . . . . . . . . . . . . . . . . . . . . . . . . . 32
6.1.4 Properties of matrices . . . . . . . . . . . . . . . . . . . . . . . . . 32
6.1.5 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
7. Inverses of Square Matrices. 34
7.1 Matrix Inverse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
7.1.1 Inverse by Gaussian elimination . . . . . . . . . . . . . . . . . . . . 35
7.2 Determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
7.2.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
7.3 Inverse using determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
7.4 Vector Cross Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
8. Eigenvalues and eigenvectors. 39
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
8.2 Eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
8.3 Decomposing Symmetric matrices . . . . . . . . . . . . . . . . . . . . . . . 44
8.4 Matrix inverse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
8.5 The Cayley-Hamilton theorem: Not examinable . . . . . . . . . . . . . . . 47
9. Integration 49
9.1 Integration : Revision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
9.1.1 Some basic integrals . . . . . . . . . . . . . . . . . . . . . . . . . . 51
9.1.2 Substitution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
9.2 Integration by parts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
10. Hyperbolic functions 54
10.1 Hyperbolic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
10.1.1 Hyperbolic functions . . . . . . . . . . . . . . . . . . . . . . . . . 56
10.1.2 More hyperbolic functions . . . . . . . . . . . . . . . . . . . . . . 57
26-Jul-2013 3
School of Mathematical Sciences Monash University
10.2 Special functions: not examinable . . . . . . . . . . . . . . . . . . . . . . 58
11. Improper integrals 60
11.1 Improper Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
11.1.1 A standard strategy . . . . . . . . . . . . . . . . . . . . . . . . . . 61
12. Comparison test for convergence 65
12.1 Comparison Test for Improper Integrals . . . . . . . . . . . . . . . . . . . 66
12.2 The General Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
13. Introduction to sequences and series. 70
13.1 Denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
13.1.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
13.1.2 Partial sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
13.1.3 Arithmetic series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
13.1.4 Fibonacci sequence . . . . . . . . . . . . . . . . . . . . . . . . . . 72
13.1.5 Geometric series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
13.1.6 Compound Interest . . . . . . . . . . . . . . . . . . . . . . . . . . 73
14. Convergence of series. 75
14.1 Innite series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
14.1.1 Convergence and divergence . . . . . . . . . . . . . . . . . . . . . 76
14.2 Tests for convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
14.2.1 Zero tail? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
14.2.2 The Comparison test . . . . . . . . . . . . . . . . . . . . . . . . . 76
15. Integral and ratio tests. 78
15.1 The Integral Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
15.2 The Ratio test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
16. Comparison test, alternating series. 83
16.1 Alternating series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
16.2 Non-positive innite series . . . . . . . . . . . . . . . . . . . . . . . . . . 85
16.3 Re-ordering an innite series . . . . . . . . . . . . . . . . . . . . . . . . . 85
17. Power series 87
17.1 Simple power series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
26-Jul-2013 4
School of Mathematical Sciences Monash University
17.2 The general power series . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
17.3 Examples of Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
17.4 Maclaurin Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
17.5 Taylor Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
17.6 Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
18. Radius of convergence 93
18.1 Radius of convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
18.2 Computing the Radius of Convergence . . . . . . . . . . . . . . . . . . . 94
18.3 Some theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
19. Function Approximation using Taylor Series 96
19.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
19.2 Taylor polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
19.3 Accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
19.4 Using Taylor series to calculate limits . . . . . . . . . . . . . . . . . . . . 102
19.5 lHopitals rule. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
20. Remainder term for Taylor series. 106
20.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
20.2 Integration by parts and Taylor series . . . . . . . . . . . . . . . . . . . . 107
21. Introduction to ODEs 111
21.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
21.2 Denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
21.3 Solution strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
21.4 General and particular solutions . . . . . . . . . . . . . . . . . . . . . . . 115
22. Separable rst order ODEs. 116
22.1 Separable equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
22.2 First order linear ODEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
22.2.1 Solving the homogeneous ODE . . . . . . . . . . . . . . . . . . . . 120
22.2.2 Finding a particular solution . . . . . . . . . . . . . . . . . . . . . 121
23. The integrating factor. 122
23.1 The Integrating Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
26-Jul-2013 5
School of Mathematical Sciences Monash University
24. Homogeneous Second order ODEs. 125
24.1 Second order linear ODEs . . . . . . . . . . . . . . . . . . . . . . . . . . 126
24.2 Homogeneous equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
25. Non-Homogeneous Second order ODEs. 131
25.1 Non-homogeneous equations . . . . . . . . . . . . . . . . . . . . . . . . . 132
25.2 Undetermined coecients . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
25.3 Exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
26. Coupled systems of ODEs 135
26.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
26.2 First method: dierentiation . . . . . . . . . . . . . . . . . . . . . . . . . 136
26.3 Second method: eigenvectors and eigenvalues . . . . . . . . . . . . . . . . 138
27. Applications of Dierential Equations 141
27.1 Applications of ODEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
27.2 Newtons law of cooling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
27.3 Pollution in swimming pools . . . . . . . . . . . . . . . . . . . . . . . . . 143
27.4 Newtonian mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
28. Functions of Several Variables 147
28.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
28.2 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
28.3 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
28.4 Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
28.5 Alternative forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
29. Partial derivatives 154
29.1 First derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
29.2 Higher derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
29.3 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
29.4 Exceptions : when derivatives do not exist . . . . . . . . . . . . . . . . . 158
30. Chain Rule, Gradient and Directional derivatives 160
30.1 The Chain Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
30.2 Gradient and Directional Derivative . . . . . . . . . . . . . . . . . . . . . 163
26-Jul-2013 6
School of Mathematical Sciences Monash University
31. Tangent planes and linear approximations 166
31.1 Tangent planes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
31.2 Linear Approximations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
32. Maxima and minima 170
32.1 Maxima and minima . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
32.2 Local extrema . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
32.3 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
32.4 Maxima, Minima or Saddle point? . . . . . . . . . . . . . . . . . . . . . . 173
26-Jul-2013 7
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
1. Vectors in 3-dimensions
School of Mathematical Sciences Monash University
1.1 Introduction
These can be dened in (at least) two ways, algebraically as objects like
v

= (1, 7, 3)
u

= (2, 1, 4)
or geometrically as arrows in space.
How can we be sure that these two denitions actually describe the same object? Equally,
how do we convert from one form to the other? That is, given (1, 2, 7) how do we draw
the arrow and likewise, given the arrow how do we extract the numbers (1, 2, 7)?
Suppose we are give two points P and Q. Suppose also that we nd the change in
coordinates from P to Q is (say) (1, 2, 7). We could also draw an arrow from P to Q.
Thus we have two ways of recording the path from P to Q, either as the numbers (1, 2, 7)
or the arrow.
Suppose now that we have another pair of points R and S and further that we nd the
change in coordinates to be (1, 2, 7). Again, we can join the points with an arrow. This
arrow will have the same direction and length as that for P to Q.
In both cases, the displacement, from start to nish, is represented by either the numbers
(1, 2, 7) or the arrow thus we can use either form to represent the vector. Note that this
means that a vector does not live at any one place in space it can be moved anywhere
provided its length and direction are unchanged.
To extract the numbers (1, 2, 7) given just the arrow simply place the arrow somewhere
in the x, y, z space, and the measure the change in coordinates from tail to tip of the
vector. Equally, to draw the vector given the numbers (1, 2, 7) is easy choose (0, 0, 0)
as the tail then the point (1, 2, 7) is the tip.
1.1.1 Notation
The components of a vector are just the numbers we use to describe the vector. In the
above, the components of v

are 1,2 and 7.


Another very very common way to write a vector, such as v

= (1, 7, 3) for example, is


v

= 1 i

+ 7j

+ 3k

. The three vectors i

, j

, k

are a simple way to remind us that the


three numbers in v

= (1, 7, 3) refer to directions parallel to the three coordinate axes


(with i

parallel to the x-axis, j

parallel to the y-axis and k

parallel to the z-axis).


In this way we can always write down any 3-dimensional vector as a linear combination
of the i

, j

, k

and thus these vectors are also known as basis vectors.


1.1.2 Algebraic properties
What rules must we observe in playing with vectors?
26-Jul-2013 9
School of Mathematical Sciences Monash University
Equality
v

= w

only when the arrows for v

and w

are identical.
Stretching
The vector v

is parallel to v

but is stretched by a factor .


Addition
To add two vectors v

and w

arrange the two so that they are tip to tail. Then


v

+ w

is the vector that starts at the rst tail and ends at the second tip.
Example 1.1
Express each of the above rules in terms of the components of vectors (i.e. in terms of
numbers like (1, 2, 7) and (a, b, c)).
Example 1.2
Given v

= (3, 4, 2) and w

= (1, 2, 3) compute v

+ w

and 2v

+ 7w

.
Example 1.3
Given v

= (1, 2, 7) draw v

, 2v

and v

.
Example 1.4
Given v

= (1, 2, 7) and w

= (3, 4, 5) draw and compute v

.
1.2 Vector Dot Product
How do we multiply vectors? We have already seen one form, stretching, v

. This
is called scalar multiplication.
Here is another form. Let v

= (v
x
, v
y
, v
z
) and w

= (w
x
, w
y
, w
z
) be a pair of vectors then
we dene the dot product v

by
v

= v
x
w
x
+ v
y
w
y
+ v
z
w
z
Example 1.5
Let v

= (1, 2, 7) and w

= (1, 3, 4). Compute v

, w

and v

What do we observe?
v

is a single number not a vector


v

= w

(v

) w

= (v

)
(a

+ b

) v

= a

+ b

The last two cases display what we call linearity.


26-Jul-2013 10
School of Mathematical Sciences Monash University
Example 1.6 : Length of a vector
Let v

= (1, 2, 7). Compute the distance from (0, 0, 0) to (1, 2, 7). Compare this with

.
We can now show that
v

= |v||w| cos
where
|v| = the length of v

=
_
v
2
x
+ v
2
y
+ v
2
z
_
1/2
|w| = the length of w

=
_
w
2
x
+ w
2
y
+ w
2
z
_
1/2
and is the angle between the two vectors.
How do we prove this? Simple start with v

and compute its length,


|v w|
2
= (v

) (v

)
= v

+ w

= |v|
2
+|w|
2
2v

and from the Cosine Rule for triangles we know


|v w|
2
= |v|
2
+|w|
2
2|v||w| cos
Thus we have
v

= |v||w| cos
This gives us a convenient way to compute the angle between any pair of vectors. If
we nd cos = 0 then we say that v

and w

are orthogonal (sometimes also called


perpendicular).
Thus v

and w

are orthogonal when v

= 0 (provided neither v

nor w

are zero).
Example 1.7
Find the angle between the vectors v

= (2, 7, 1) and w

= (3, 4, 2)
1.2.1 Unit Vectors
A vector is said to be a unit vector if its length is one. That is, v

is a unit vector when


v

= 1.
26-Jul-2013 11
School of Mathematical Sciences Monash University
1.3 Vector Cross Product
This is another way to multiply vectors. Start with v

= (v
x
, v
y
, v
z
) and w

= (w
x
, w
y
, w
z
).
Then we dene the cross product v

by
v

= (v
y
w
z
v
z
w
y
, v
z
w
x
v
x
w
z
, v
x
w
y
v
y
w
x
)
From this denition we observe
v

is a vector
v

= w

= 0

(v

) w

= (v

)
(a

+ b

) v

= a

+ b

(v

) v

= (v

) w

= 0

Example 1.8
Verify all of the above.
Example 1.9
Given v

= (1, 2, 7) and w

= (2, 3, 5) compute v

, and its dot product with each of


v

and w

.
1.3.1 Interpreting the cross product
We know that v

is a vector and we know how to compute it. But can we describe


this vector? First we need a vector, so lets assume that v

= 0

. Then what can we


say about the direction and length of v

?
The rst thing we should note is that the cross product is a vector which is orthogonal
to both of the original vectors. Thus v

is a vector that is orthogonal to v

and to
w

. This fact follows from the denition of the cross product.


Thus we must have
v

= n

where n

is a unit vector orthogonal to both v

and w

and is some unknown number


(at this stage).
How do we construct n

and ? Lets do it!


26-Jul-2013 12
School of Mathematical Sciences Monash University
1.3.2 Right Hand Thumb rule
For any choice of v

and w

you can see that there are two choices for n

one points in
the opposite direction to the other. Which one do we choose? Its up to us to make a
hard rule. This is it. Place your right hand palm so that your ngers curl over from v
to w

. Your thumb then points in the direction of v

.
Now for , we will show that
|v

| = = |v||w| sin
How? First we build a triangle from v

and w

and then compute the cross product for


each pair of vectors
v

(v

) v

(v

) w

(one for each of the three vertices). We need to compute each .


Now since (v

) w

= (v

) for any number we must have

in v

n
proportional to |v||w|, likewise for the other s. Thus

= |v||w|

= |v||v w|

= |w||v w|

where each depends only on the angle between the two vectors on which it was built
(i.e.

depends only on the angle between v

and v

).
But we also have v

= (v

) v

= (v

) w

which implies that

which in turn gives us

|v w|
=

|w|
=

|v|
(Were in the home straight...)
But we also have the Sine Rule for triangles
sin
|v w|
=
sin
|w|
=
sin
|v|
and so

= k sin ,

= k sin ,

= k sin
where k is a pure number that does not depend on any of the angles nor on any of
lengths of the edges the value of k is the same for every triangle. We can choose a
trivial case to compute k, simply put v

= (1, 0, 0) and w

= (0, 1, 0). Then we nd k = 1.


Its been a merry ride but weve found that
|v

| = |v||w| sin
26-Jul-2013 13
School of Mathematical Sciences Monash University
Example 1.10
Show that |v

| also equals the area of the parallelogram formed by v

and w

.
Vector Dot and Cross products
Let v

= (v
x
, v
y
, v
z
) and w

= (w
x
, w
y
, w
z
). Then the Dot Product of v

and w

is
dened by
v

= v
x
w
x
+ v
y
w
y
+ v
z
v
z
.
while the Cross Product is dened by
v

= (v
y
w
z
v
z
w
y
, v
z
w
x
v
x
w
z
, v
x
w
y
v
y
w
x
)
1.4 Scalar and Vector projections
These are like shadows and there are two basic types, scalar and vector projections.
1.4.1 Scalar Projections
This is simply the shadow cast by one vector on another.
Example 1.11
What is the length (i.e. scalar projection) of v

= (1, 2, 7) in the direction of the vector


w

= (2, 3, 4)?
Scalar projection
The scalar projection, v
w
, of v

in the direction of w

is given by
v
w
=
v

|w|
1.4.2 Vector Projection
This time we produce a vector shadow with length equal to the scalar projection.
26-Jul-2013 14
School of Mathematical Sciences Monash University
Example 1.12
Find the vector projection of v

= (1, 2, 7) in the direction of w

= (2, 3, 4)
Vector projection
The vector projection, v

w
, of v

in the direction of w

is given by
v

w
=
_
v

|w|
2
_
w

Example 1.13
Given v

= (1, 2, 7) and w

= (2, 3, 4) express v

in terms of w

and a vector perpendicular


to w

.
This example shows how a vector may be resolved into its parts parallel and perpendic-
ular to another vector.
26-Jul-2013 15
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
2. Three-Dimensional Euclidean Geometry. Lines.
School of Mathematical Sciences Monash University
2.1 Lines in 3-dimensional space
Through any pair of distinct points we can always construct a straight line. These lines
are normally drawn to be innitely long in both directions.
Example 2.1
Find all points on the line joining (2, 4, 0) and (2, 4, 7)
Example 2.2
Find all points on the line joining (2, 0, 0) and (2, 4, 7)
These equations for the line are all of the form
x(t) = a + pt , y(t) = b + qt , z(t) = c + rt
where t is a parameter (it selects each point on the line) and the numbers a, b, c, p, q, r
are computed from the coordinates of two points on the line. (There are other ways to
write an equation for a line.)
How do we compute a, b, c, p, q, r? First put t = 0, then x = a, y = b, z = c. That is
(a, b, c) are the coordinates of one point on the line and so a, b, c are known. Next, put
t = 1, then x = a +p, y = b +q, z = c +r. Take this to be the second point on the line,
and thus solve for p, q, r.
A common interpretation is that (a, b, c) are the coordinates of one (any) point on the
line and (p, q, r) are the components of a (any) vector parallel to the line.
Example 2.3
Find the equation of the line joining the two points (1, 7, 3) and (2, 0, 3).
Example 2.4
Show that a line may also be expressed as
x a
p
=
y b
q
=
z c
r
provided p = 0, q = 0 and r = 0. This is known as the Symmetric Form of the equation
for a a straight line.
Example 2.5
In some cases you may nd a small problem with the form suggested in the previous
example. What is that problem and how would you deal with it?
Example 2.6
Determine if the line dened by the points (1, 0, 1) and (1, 2, 0) intersects with the line
dened by the points (3, 1, 0) and (1, 2, 5).
26-Jul-2013 17
School of Mathematical Sciences Monash University
Example 2.7
Is the line dened by the points (3, 7, 1) and (2, 2, 1) parallel to the line dened by
the points (1, 4, 1) and (0, 5, 1).
Example 2.8
Is the line dened by the points (3, 7, 1) and (2, 2, 1) parallel to the line dened by
the points (1, 4, 1) and (2, 23, 5).
2.2 Vector equation of a line
The parametric equations of a line are
x(t) = a + pt , y(t) = b + qt z(t) = c + rt
Note that
(a, b, c) = the vector to one point on the line
(p, q, r) = the vector from the rst point to
the second point on the line
= a vector parallel to the line
Lets put d

= (a, b, c), v

= (p, q, r) and r

(t) = (x(t), y(t), z(t)), then


r

(t) = d

+ tv

This is known as the vector equation of a line.


Example 2.9
Write down the vector equation of the line that passes through the points (1, 2, 7) and
(2, 3, 4).
Example 2.10
Write down the vector equation of the line that passes through the points (2, 3, 7) and
(4, 1, 2).
Example 2.11
Find the shortest distance between the pair of lines described in the two previous ex-
amples. Hint : Find any vector that joins a point from one line to the other and then
compute the scalar projection of this vector onto the vector orthogonal to both lines (it
helps to draw a diagram).
26-Jul-2013 18
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
3. Three-Dimensional Euclidean Geometry. Planes.
School of Mathematical Sciences Monash University
3.1 Planes in 3-dimensional space
A plane in 3-dimensional space is a at 2-dimensional surface. The standard equation
for a plane in 3-d is
ax + by + cz = d
where a, b, c and d are some bunch of numbers that identify this plane from all other
planes. (There are other ways to write an equation for a plane, as we shall see).
Example 3.1
Sketch each of the planes z = 1, y = 3 and x = 1.
3.1.1 Constructing the equation of a plane
A plane is uniquely determined by any three points (provided not all three points are
contained on a line). Recall, that a line is fully determined by any pair of points on the
line.
Lets nd the equation of the plane that passes through the three points (1, 0, 0), (0, 3, 0)
and (0, 0, 2). Our game is to compute a, b, c and d. We do this by substituting each point
into the above equation,
1st point a 1 + b 0 + c 0 = d
2nd point a 0 + b 3 + c 0 = d
3rd point a 0 + b 0 + c 2 = d
Now we have a slight problem, we are trying to compute 4 numbers, a, b, c, d but we
only have 3 equations. We have to make an arbitrary choice for one of the 4 numbers
a, b, c, d. Lets set d = 6. Then we nd from the above that a = 6, b = 2 and c = 3.
Thus the equation of the plane is
6x + 2y + 3z = 6
Example 3.2
What equation do you get if you chose d = 1 in the previous example? What happens
if you chose d = 0?
Example 3.3
Find an equation of the plane that passes through the three points (1, 0, 0), (1, 2, 0)
and (2, 1, 5).
26-Jul-2013 20
School of Mathematical Sciences Monash University
3.1.2 Parametric equations for a plane
Recall that a line could be written in the parametric form
x(t) = a + pt
y(t) = b + qt
z(t) = c + rt
A line is 1-dimensional so its points can be selected by a single parameter t.
However, a plane is 2-dimensional and so we need two parameters (say u and v) to select
each point. Thus its no surprise that every plane can also be described by the following
equations
x(u, v) = a + pu + lv
y(u, v) = b + qu + mv
z(u, v) = c + ru + nv
Now we have 9 parameters a, b, c, p, q, r, l, m and n. These can be computed from the
coordinates of three (distinct) points on the plane. For the rst point put (u, v) = (0, 0),
the second put (u, v) = (1, 0) and for the nal point put (u, v) = (0, 1). Then solve for
a through to n (its easy!).
Example 3.4
Find the parametric equations of the plane that passes through the three points (1, 0, 0),
(1, 2, 0) and (2, 1, 5).
Example 3.5
Show that the parametric equations found in the previous example describe exactly the
same plane as found in Example 3.3 (Hint : substitute the answers from Example 3.4
into the equation found in Example 3.3).
Example 3.6
Find the parametric equations of the plane that passes through the three points (1, 2, 1),
(1, 2, 3) and (2, 1, 5).
Example 3.7
Repeat the previous example but with points re-arranged as (1, 2, 1), (2, 1, 5) and
(1, 2, 3). You will nd that the parametric equations look dierent yet you know they
describe the same plane. If you did not know this last fact, how would you prove that
the two sets of parametric equations describe the same plane?
26-Jul-2013 21
School of Mathematical Sciences Monash University
3.2 Vector equation of a plane
The Cartesian equation for a plane is
ax + by + cz = d
for some bunch of numbers a, b, c and d. We will now re-express this in a vector form.
Suppose we know one point on the plane, say (x, y, z) = (x, y, z)
0
, then
ax
0
+ by
0
+ cz
0
= d
a(x x
0
) + b(y y
0
) + c(z z
0
) = 0
This is an equivalent form of the above equation.
Now suppose we have two more points on the plane (x, y, z)
1
and (x, y, z)
2
. Then
a(x
1
x
0
) + b(y
1
y
0
) + c(z
1
z
0
) = 0
a(x
2
x
0
) + b(y
2
y
0
) + c(z
2
z
0
) = 0
Put x

10
= (x
1
x
0
, y
1
y
0
, z
1
z
0
) and x

20
= (x
2
x
0
, y
2
y
0
, z
2
z
0
). Notice that
both of these vectors lie in the plane and that
(a, b, c) x

10
= (a, b, c) x

20
= 0
What does this tell us? Simply that both vectors are orthogonal to the vector (a, b, c).
Thus we must have that
(a, b, c) = the normal vector to the plane
Now lets put
n

= (a, b, c) = the normal vector to the plane


d

= (x
0
, y
0
, z
0
) = one (any) point on the plane
r

= (x, y, z) = a typical point on the plane


Then we have
n

(r

) = 0
This is the vector equation of a plane.
Example 3.8
Find the vector equation of the plane that contains the points (1, 2, 7), (2, 3, 4) and
(1, 2, 1).
Example 3.9
Re-express the previous result in the form ax + by + cz = d.
26-Jul-2013 22
School of Mathematical Sciences Monash University
Example 3.10
Find the shortest distance between the pair of planes 2x+3y4z = 2 and 4x+6y8z = 3.
An investment rm is hiring mathematicians. After the rst round of in-
terviews, three hopeful recent graduatesa pure mathematician, an applied
mathematician, and a graduate in mathematical nanceare asked what
starting salary they are expecting. The pure mathematician: Would $30,000
be too much? The applied mathematician: I think $60,000 would be OK.
The maths nance person: What about $300,000? The personnel ocer is
abbergasted: Do you know that we have a graduate in pure mathematics
who is willing to do the same work for a tenth of what you are demanding!?
Well, I thought of $135,000 for me, $135,000 for you - and $30,000 for the
pure mathematician who will do the work.
26-Jul-2013 23
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
4. Linear systems of equations
School of Mathematical Sciences Monash University
4.1 Examples of Linear Systems
4.1.1 Bags of coins
We have three bags with a mixture of gold, silver and copper coins. We are given the
following information
Bag 1 contains 10 gold, 3silver, 1 copper and weighs 60g
Bag 2 contains 5 gold, 1 silver and 2 copper and weighs 30g
Bag 3 contains 3 gold, 2silver, 4 copper and weighs 25g
The question is What are the respective weights of the Gold, Silver and Copper coins?
Let G, S and C denote the weight of each of the gold, silver and copper coins. Then we
have the system of equations
10G + 3S + C = 60
5G + S + 2C = 30
3G + 2S + 4C = 25
4.1.2 Silly puzzles
John and Marys ages add to 75 years. When John was half his present age John was
twice as old as Mary. How old are they?
We have just two equations,
J + M = 75
1
2
J 2M = 0
4.1.3 Intersections of planes
Its easy to imagine three planes in space. Is it possible that they share one point in
common? Here are the equations for three such planes
3x + 7y 2z = 0
6x + 16y 3z = 1
3x + 9y + 3z = 3
Can we solve this system for (x, y, z)?
In all of the above examples we need to unscramble the set of linear equations to extract
the unknowns (e.g. G, S, C etc.).
26-Jul-2013 25
School of Mathematical Sciences Monash University
4.2 A standard strategy
We start with the previous example
3x + 7y 2z = 0 (1)
6x + 16y 3z = 1 (2)
3x + 9y + 3z = 3 (3)
Suppose by some process we were able to rearrange these equations into the following
form
3x + 7y 2z = 0 (1)
2y + z = 1 (2)

4z = 4 (3)

Then we could solve (3)

for z
(3)

4z = 4 z = 1
and then substitute into (2)

to solve for y
(2)

2y + 1 = 1 y = 1
and substitute into (1) to solve for x
(1) 3x 7 2 = 0 x = 3
The question is : How do we get the modied equations (1), (2)

and (3)

?
The general trick is to take suitable combinations of the equations so that we can elim-
inate various terms. The trick is applied as many times as we need to turn the original
equations into the simple form like (1), (2)

and (3)

.
Lets start with the rst pair of the original equations
3x + 7y 2z = 0 (1)
6x + 16y 3z = 1 (2)
We can eliminate the 6x in equations (2) by replacing equation (2) with (2) 2(1),
0x + (16 14)y + (3 + 4)z = 1 (2)

2y + z = 1 (2)

Likewise, for the 3x term in equation (3) we replace equation (3) with (3) (1),
2y + 5z = 3 (3)

26-Jul-2013 26
School of Mathematical Sciences Monash University
At this point our system of equations is
3x + 7y 2z = 0 (1)
2y + z = 1 (2)

2y + 5z = 3 (3)

The last step is to eliminate the 2y term in the last equation. We do this by replacing
equation (3)

with (3)

(2)

4z = 4 (3)

So nally we arrive at the system of equations


3x + 7y 2z = 0 (1)
2y + z = 1 (2)

4z = 4 (3)

which, as before, we solve to nd z = 1, y = 1 and x = 3.


The procedure we just went through is known as a reduction to upper triangular form
and we used elementary row operations to do so. We then solved for the unknowns by
back substitution.
This procedure is applicable to any system of linear equations (though beware, for some
systems the back substitution method requires special care, well see examples later).
The general strategy is to eliminate all terms below the main diagonal, working column
by column from left to right.
4.3 Lines and planes
In previous lecture we saw how we could construct the equations for lines and planes.
Now we can answer some simple questions.
How do we compute the intersection between a line and a plane? Can we be sure that
they do intersect? And what about the intersection of a pair or more of planes?
The general approach to all of these questions is simply to write down equations for each
of the lines and planes and then to search for a common point (i.e. a consistent solution
to the system of equations).
Example 4.1
Find the intersection of the plane y = 0 with the plane 2x + 3y 4z = 1.
Example 4.2
Find the intersection of the line x(t) = 1 + 3t, y(t) = 3 2t, z(t) = 1 t with the plane
2x + 3y 4z = 1.
26-Jul-2013 27
School of Mathematical Sciences Monash University
Example 4.3
Find the intersection of the three planes 2x + 3y z = 1, x y = 2 and x = 1
In general, three planes may intersect at a single point or along a common line or even
not at all.
Here are some examples (there are others) of how planes may (or may not) intersect.
No point of intersection
One point of intersection
Intersection in a common line
26-Jul-2013 28
School of Mathematical Sciences Monash University
Example 4.4
What other examples can you draw of intersecting planes?
Three men are in a hot-air balloon. Soon, they nd themselves lost in a
canyon somewhere. One of the three men says, Ive got an idea. We can
call for help in this canyon and the echo will carry our voices far.
So he leans over the basket and yells out, Helllloooooo! Where are we?
(They hear the echo several times).
15 minutes later, they hear this echoing voice: Helllloooooo! Youre lost!!
One of the men says, That must have been a mathematician. Puzzled,
one of the other men asks, Why do you say that? The reply: For three
reasons.
(1) he took a long time to answer,
(2) he was absolutely correct, and
(3) his answer was absolutely useless.
26-Jul-2013 29
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
5. Gaussian Elimination
School of Mathematical Sciences Monash University
5.1 Gaussian elimination and back-substitution
Example 5.1 : Typical layout
2x + 3y + z = 10
x + 2y + 2z = 10
4x + 8y + 11z = 49
(1)
(2)

2(2) (1)
(3)

(3) 2(1)
2x + 3y + z = 10
y + 3z = 10
2y + 9z = 29
(1)
(2)

(3)

(3)

2(2)

2x + 3y + z = 10
y + 3z = 10
3z = 9
(1)
(2)

(3)

Now we solve this system using back-substitution, z = 3, y = 1, x = 2.


Note how we record the next set of row-operations on each equation. This makes it
much easier for someone else to see what you are doing and it also helps you track down
any arithmetic errors.
5.2 Gaussian elimination
In the previous example we found
2x + 3y + z = 10
y + 3z = 10
3z = 9
(1)
(2)

(3)

Why stop there? We can apply more row-operations to eliminate terms above the
diagonal. This does not involve back-substitution. This method is known as Gaussian
elimination. Take note of the dierence!
Example 5.2
Continue from the previous example and use row-operations to eliminate the terms above
the diagonal. Hence solve the system of equations.
26-Jul-2013 31
School of Mathematical Sciences Monash University
5.2.1 Gaussian elimination strategy
1. Use row-operations to eliminate elements below the diagonal.
2. Use row-operations to eliminate elements above the diagonal.
3. If possible, re-scale each equation so that each diagonal element = 1.
4. The right hand side is now the solution of the system of equations.
If you bail out after step 1 you are doing Gaussian elimination with back-substitution
(this is usually the easier option).
5.3 Exceptions
Here are some examples where problems arise.
Example 5.3 : A zero on the diagonal
2x + y + 2z + w = 2
2x + y z + 2w = 1
x 2y + z w = 2
x + 3y z + 2w = 2
(1)
(2)

(2) (1)
(3)

2(3) (1)
(4)

2(4) (1)
2x + y + 2z + w = 2
0y 3z + w = 1
5y + 0z 3w = 6
+ 5y 4z + 3w = 2
(1)
(2)

(3)

(3)

(2)

(4)

The zero on the diagonal on the second equation is a serious problem, it means we can
not use that row to eliminate the elements below the diagonal term. Hence we swap the
second row with any other lower row so that we get a non-zero term on the diagonal.
Then we proceed as usual. The result is w = 2, z = 1, y = 0 and x = 1.
Example 5.4
Complete the above example.
26-Jul-2013 32
School of Mathematical Sciences Monash University
Example 5.5 : A consistent and under-determined system
Suppose we start with three equations and we wind up with
2x + 3y z = 1
5y + 5z = 1
0z = 0
(1)
(2)

(3)

The last equation tells us nothing! We cant solve it for any of x, y and z. We really only
have 2 equations, not 3. That is 2 equations for 3 unknowns. This is an under-determined
system.
We solve the system by choosing any number for one of the unknowns. Say we put z =
where is any number (our choice). Then we can leap back into the equations and use
back-substitution.
The result is a one-parameter family of solutions
x =
1
5
, y =
1
5
+ , z =
Since we found a solution we say that the system is consistent.
Example 5.6 : An inconsistent system
Had we started with
2x + 3y z = 1
x y + 2z = 0
3x + 2y + z = 0
(1)
(2)
(3)
we would have arrived at
2x + 3y z = 1
5y + 5z = 1
0z = 2
(1)
(2)

(3)

This last equation makes no sense as there are no nite values for z such that 0z = 2
and thus we say that this system is inconsistent and that the system has no solution.
26-Jul-2013 33
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
6. Matrices
School of Mathematical Sciences Monash University
6.1 Matrices
When we use row-operations on systems of equations such as
3x + 2y z = 3
x y + z = 1
2x + y z = 0
the x, y, z just hang around. All the action occurs on the coecients and the right hand
side. To assist in the bookkeeping we introduce a new notation, matrices,
_
_
3 2 1
1 1 1
2 1 1
_
_
_
_
x
y
x
_
_
=
_
_
3
1
0
_
_
Each [ ] is a matrix,
_
_
3 2 1
1 1 1
2 1 1
_
_
is a square 33 matrix, while
_
_
x
y
z
_
_
and
_
_
3
1
0
_
_
are 1-dimensional matrices (also called column vectors).
We can recover the original system of equations by dening a rule for multiplying ma-
trices,
_

_


a b c d


_

_
_

_
e
f
g
h

.
.
.
_

_
=
_

_


i


.
.
.
_

_
i = a e + b f + c g + d h +
Example 6.1
Write the above system of equations in matrix form.
_
_
3 2 1
1 1 1
2 1 1
_
_
_
_
x
y
z
_
_
=
_
_
3 x + 2 y 1 z
1 x 1 y + 1 z
2 x + 1 y 1 z
_
_
26-Jul-2013 35
School of Mathematical Sciences Monash University
Example 6.2
Compute
_
2 3
4 1
_ _
1 7
0 2
_
and
_
1 7
0 2
_ _
2 3
4 1
_
Note that we can only multiply matrices that t together. That is, if A and B are a pair
of matrices then in order that AB makes sense we must have the number of columns of
A equal to the number of rows of B.
Example 6.3
Does the following make sense?
_
2 3
4 1
_
_
_
1 7
0 2
4 1
_
_
6.1.1 Notation
We use capital letters to represent matrices,
A =
_
_
3 2 1
1 1 1
2 1 1
_
_
, X =
_
_
x
y
x
_
_
, B =
_
_
3
1
0
_
_
and our previous system of equations can then be written as
AX = B
Entries within a matrix are denoted by subscripted lowercase letters. Thus for the matrix
B above we have b
1
= 3, b
2
= 1 and b
3
= 0 while for the matrix A we have
A =
_
_
3 2 1
1 1 1
2 1 1
_
_
=
_
_
a
11
a
12
a
13
a
21
a
22
a
23
a
31
a
32
a
33
_
_
a
ij
= the entry in row i and column j of A
To remind us that A is a square matrix with elements a
ij
we sometimes write A = [a
ij
].
6.1.2 Operations on matrices
Equality:
A = B
only when all entries in A equal those in B.
Addition: Normal addition of corresponding elements.
26-Jul-2013 36
School of Mathematical Sciences Monash University
Multiplication by a number: A = times each entry of A
Multiplication of matrices:
_

_

_

_
_

_
=
_

_
Transpose: Flip rows and columns, denoted by [ ]
T
.
_
1 2 7
0 3 4
_
T
=
_
_
1 0
2 3
7 4
_
_
6.1.3 Some special matrices
The Identity matrix:
I =
_

_
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
_

_
For any square matrix A we have IA = AI = A.
The Zero matrix: A matrix full of zeroes!
Symmetric matrices: Any matrix A for which A = A
T
.
Skew-symmetric matrices: Any matrix A for which A = A
T
. Sometimes also
called anti-symmetric.
6.1.4 Properties of matrices
AB = BA
(AB)C = A(BC)
(A
T
)
T
= A
(AB)
T
= B
T
A
T
26-Jul-2013 37
School of Mathematical Sciences Monash University
6.1.5 Notation
For the system of equations
3x + 2y z = 1
x y + z = 4
2x + y z = 1
we call
_
_
3 2 1
1 1 1
2 1 1
_
_
the coecient matrix and
_
_
3 2 1 1
1 1 1 4
2 1 1 1
_
_
the augmented matrix.
When we do row-operations on a system we are manipulating the augmented matrix.
But each incarnation represents a system of equations for the same original values for
x, y and z. Thus if A and A

are two augmented matrices for the same system, then we


write
A A

The squiggle means that even though A and A

are not the same matrices, they do give


us the same values for x, y and z.
Example 6.4
Solve the system of equations
3x + 2y z = 1
x y + z = 4
2x + y z = 1
using matrix notation.
An accountant is someone who is good with numbers but lacks the personality
to be a statistician.
26-Jul-2013 38
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
7. Inverses of Square Matrices.
School of Mathematical Sciences Monash University
7.1 Matrix Inverse
Suppose we have a system of equations
_
a b
c d
_ _
x
y
_
=
_
u
v
_
and that we write in the matrix form
AX = B
Can we nd another matrix, call it A
1
, such that
A
1
A = I = the identity matrix
If so, then we have
A
1
AX = A
1
B X = A
1
B
Thus we have found the solution of the original system of equations.
For a 2 2 matrix it is easy to verify that
A
1
=
_
a b
c d
_
1
=
1
ad bc
_
d b
c a
_
But how do we compute the inverse A
1
for other (square) matrices?
Here is one method.
7.1.1 Inverse by Gaussian elimination
Use row-operations to reduce A to the identity matrix.
Apply exactly the same row-operations to a matrix set initially to the identity.
The nal matrix is the inverse of A.
We usually record this process in a large augmented matrix.
Start with [A|I].
Apply row operations to obtain [I|A
1
]
Crack open the champagne.
26-Jul-2013 40
School of Mathematical Sciences Monash University
Example 7.1
Find the inverse for A =
_
1 7
3 4
_
Note that not all matrices will have an inverse. For example, if
A =
_
a b
c d
_
then
A
1
=
1
ad bc
_
d b
c a
_
and for this to be possible we must have ad bc = 0.
We call this magic number the determinant of A. If it is zero then A does not have an
inverse.
The question is is there a similar rule for an N N matrix? That is, a rule which can
identify those matrices which have an inverse.
7.2 Determinants
The denition is a bit involved, here it is.
For a 2 2 matrix A =
_
a b
c d
_
dene det A = ad bc.
For an NN matrix A create a sub-matrix S
ij
of A by deleting row I and column
J.
Then dene
det A = a
11
det S
11
a
12
det S
12
+ a
13
det S
13
a
1N
det S
1N
Thus to compute det A you have to compute a chain of determinants, from (N 1)
(N 1) determinants all the way down to 2 2 determinants. This is tedious and very
prone to arithmetic errors!
Note the alternating plus minus signs, its very important!!!
7.2.1 Notation
We often write det A = |A|.
26-Jul-2013 41
School of Mathematical Sciences Monash University
Example 7.2
Compute the determinant of
A =
_
_
1 7 2
3 4 5
6 0 9
_
_
We can also expand the determinant about any row or column provided we observe the
following pattern of signs.
_

_
+ + +
+ + +
+ + +
+ + +
_

_
Example 7.3
By expanding about the second row compute the determinant of
A =
_
_
1 7 2
3 4 5
6 0 9
_
_
Example 7.4
Compute the determinant of
A =
_
_
1 2 7
0 0 3
1 2 1
_
_
7.3 Inverse using determinants
Here is another way to compute the inverse matrix.
Select a row I and column J of A.
Compute (1)
i+j det SIJ
det A
Store this at row J and column I in the inverse matrix.
Repeat for all other entries in A.
That is , if
A = [ a
IJ
]
then
A
1
=
1
det A
_
(1)
I+J
det S
JI

This method for the inverse works but it is rather tedious.


The best way is to compute the inverse by Gaussian elimination, i.e. [A|I] [I|A
1
].
26-Jul-2013 42
School of Mathematical Sciences Monash University
7.4 Vector Cross Products
The rule for a vector cross product can be conveniently expressed as a determinant.
Thus if v

= v
x
i

+ v
y
j

+ v
z
k

and w

= w
x
i

+ w
y
j

+ w
z
k

then
v

k
v
x
v
y
v
z
w
x
w
y
w
z

A graduate student from Trinity


Computed the cube of innity;
But it gave him the dgets
To write down all those digits,
So he dropped maths and took up divinity.
26-Jul-2013 43
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
8. Eigenvalues and eigenvectors.
School of Mathematical Sciences Monash University
8.1 Introduction
Okay, its late in the aftrenoon, were feeling a little sleepy and we need somthing to get
our minds red up. So we play a little game. We start with this simple 3 3matrix
R =
_
_
1 2 0
2 1 0
0 0 3
_
_
and when we apply R to any vector of the form v = [0, 0, 1]
T
we observe the curious fact
that the vector remains unchanged apart from an overall scaling by 3. That is
Rv = 3v
Now we are wide awake and ready to play this game at full speed. Qustions that come
to mind would (should) include,
Can we nd such a vector for any matrix?
How many distinct vectors are there?
Can we nd vectors like v but with a dierent scaling?
This is a simple example of what is known as an eigenvector equation. The key feature
is that the action of the matrix on the vector produces a new vector that is parallel to
the original vector (and in our case, it also happens to be 3 times as long).
Eigenvalues and eigenvectors
If A is square matrix and v is a column vector v with
Av = v
for some non-zero vector v then we say that the matrix A has v as an eigenvector
with eigenvalue .
For the example of the 3 3 matrix given above we have an eigenvalue equal to 3 and
an eigenvector of the form v = [0, 0, 1]
T
.
Example 8.1
Show that v = [8, 1]
T
is an eigenvector of the matrix A =
_
6 16
1 4
_
Example 8.2
The matrix in the previous example has a second eigenvector this time with the eigen-
value -2. Find that eigenvector.
26-Jul-2013 45
School of Mathematical Sciences Monash University
Example 8.3
Let v
1
and v
2
be two eigenvectors of some matrix. Is it possible to choose and so
that v
1
+ v
2
is also an eigenvector?
Now we can reexpress our earlier questions as follows.
Does every matrix possess an eigenvector?
How many eigenvalues can a matrix have?
How do we compute the eigenvalues?
Is this just pretty mathematics or is there a point to this game?
Good questions indeed. Lets see what we make of them. We will start with the issue
of constructing the eigenvalues (assuming, for the moment, that they exist).
8.2 Eigenvalues
Our game here is to nd the non-zero values of , if any, that allows the equation
Av = v
to have non-zero solutions for v. Take that is given, then re-arrange the equation to
(A I) v = 0
where I is the identity matrix (of the same shape as A). Since we are chasing non-zero
solutions for v we must have the determinant of A I equal to zero. That is, we
require that 0 = det(A I). This is a polynomial equation in and is known as the
characteristic equation for .
Characteristic equation
The eigenvalues of a matrix A are solutions of the polynomial equation
0 = det(A I)
This is called the characteristic equation of A. If A is an N N matrix, then this
equation will be a polynomial of degree N in . The eigenvalues may in general be
complex numbers.
26-Jul-2013 46
School of Mathematical Sciences Monash University
Example 8.4
Compute both eigenvalues of A =
_
6 16
1 4
_
We can now answer the pervious question How many eigenvalues can we nd for a given
matrix? If A is an NN matrix then the characteristic equation will be a polynomial of
degree N and so we can expect at most N distinct eigenvalues (one for each root). The
keyword here is distinct it is possible that the characteristic equation has repeated
roots. In such cases we will nd less than N (distinct) eigenvalues, as shown in the
following example.
Example 8.5
Show that the matrix A =
_
1 3
0 1
_
has only one eigenvalue.
Example 8.6
Look carefully at the previous matrix. It describes a stretch along the x-axis. Use this
fact to argue that the matrix can have only one eigenvalue. This is a pure geometrical
argument, you should not need to to do any calculations.
Example 8.7 A characteristic equation
Show that the characteristic equation for the matrix
A =
_
_
5 8 16
4 1 8
4 4 11
_
_
is given by
0 =
3
+ 5
2
+ 3 9
Example 8.8 The eigenvalues
Show that the eigenvalues of the previous example are = 1 and = 3 (this is a
double root of the characteristic equation).
Example 8.9 Simple eigenvalue
We now know that matrix
A =
_
_
5 8 16
4 1 8
4 4 11
_
_
has an eigenvalue equal to 1 (and two others which we will deal with in the next example).
How do we compute the eigenvector? We return to the eigenvector equation with = 1,
that is
_
_
5 8 16
4 1 8
4 4 11
_
_
_
_
a
b
c
_
_
=
_
_
a
b
c
_
_
26-Jul-2013 47
School of Mathematical Sciences Monash University
in which the [a, b, c]
T
is the eigenvector. We can make our job a little bit tidier by
shifting everything to the left hand side.
_
_
4 8 16
4 0 8
4 4 12
_
_
_
_
a
b
c
_
_
= 0
Our game now is to solve these equations for a, b and c. This we can do using Gaussian
elimination. After the rst stage, where we eliminate the lower triangular part, we obtain
_
_
4 8 16
0 8 8
0 0 0
_
_
_
_
a
b
c
_
_
= 0
Note that the last row is full of zeros. Are we surprised? No. Why Not? Well, since we
were told that the matrix A has = 1 as an eigenvalue we also know that det(A1I) = 0
which in turn tells us that at least one of the rows of A 1I must be a (hidden)
linear combination of the other rows (and Gaussian elimination reveals that hidden
combination). So seeing a row of zeros is conrmation that we have det(A 1I) = 0.
Now lets return to the matter of solving the equations. Using back-substitution we nd
that every solution is of the form
_
_
a
b
c
_
_
=
_
_
2
1
1
_
_
where is any number. We can set = 1 and this will give us a typical eigenvector
for the eigenvalue = 1. All other eigenvectors, for this eigenvalue, are parallel to this
eigenvector (diering only in length). Is that what we expected, that there would be an
innite set of eigenvectors for a given eigenvalue? Yes just look back at the denition,
Av = v. If v is a solution of this equation then so too is v. This is exactly what we
have just found.
Example 8.10 A double eigenvalue
Now lets nd the eigenvectors corresponding to = 3. We start with
_
_
8 8 16
4 4 8
4 4 8
_
_
_
_
a
b
c
_
_
= 0
After doing our Gaussian elimination we nd
_
_
8 8 16
0 0 0
0 0 0
_
_
_
_
a
b
c
_
_
= 0
This time we nd that we have two rows of zeros. This is not a surprise (agreed?)
because we know that = 3 is a double root of the characteristic equation. With two
rows of zeros we are forced to introduce two free parameters, say and , leading to
_
_
a
b
c
_
_
=
_
_
2

_
_
=
_
_
1
1
0
_
_
+
_
_
2
0
1
_
_
26-Jul-2013 48
School of Mathematical Sciences Monash University
This shows that every eigenvector for = 3 is a linear combination of the pair of
vectors [1, 1, 0]
T
and [2, 0, 1]
T
.
Example 8.11
Show that eigenvectors of the previous example can also be constructed from linear
combinations of [1, 1, 0]
T
and [1, 1, 1]
T
.
8.3 Decomposing Symmetric matrices
Earlier on we asked what is the point of computing eigenvectors and eigenvalues (other
than pure fun)? Here we will develop some really nice results that follow once we know
the eigenvalues and eigenvectors. Though many of the results we are about to explore
also apply to general square matrices they are much easier to present (and prove) for
real symmetric matrices that posses a complete set of eigenvalues (i.e. no multiple roots
in the characteristic equation). This restriction is not so severe as to be meaningless for
many of the matrices encountered in mathematical physics (and other elds) are often
of this class.
Real symmetric matrices with complete eigenvalues
If A is an N N real symmetric matrix with N distinct eigenvalues
i
, i =
1, 2, 3 N with corresponding eigenvectors v
i
, i = 1, 2, 3 N then
The eigenvalues are real,
i
=

i
, i = 1, 2, 3 N and
The eigenvectors for distinct eigenvalues are orthogonal, v
T
i
v
j
= 0, i = j.
We will only prove the rst of these theorems, the second is left as an example for you
to play with (it is not all that hard).
We start by constructing v
T
Av (where the bar over the v means complex conjugation).
This is just one number, that is a 1 1 matrix. Thus it equals its own transpose. So we
have
v
T
Av =
_
v
T
Av
_
T
now use (BC)
T
= (CB)
T
= (Av)
T
v and again
= v
T
A
T
v but A
T
= A
= v
T
A v
Now from the denition Av = v we also have, by taking complex conjugates and noting
that A is real, A v =

v. Substitute this into the previous equation to obtain
v
T
Av = v
T

v =

v
T
v
26-Jul-2013 49
School of Mathematical Sciences Monash University
But look now at the left hand side. We can manipulate this as follows
v
T
Av = v
T
(Av)
= v
T
v
= v
T
v
Compare this with our previous equation and you will see that we must have
v
T
v =

v
T
v
Finally we notice that v
T
v = v
T
v = v
2
1
+ v
2
2
+ v
2
3
v
2
N
= 0. So this leaves just

=
Our job is done, we have proved that the eigenvalue must be real.
Now here comes a very nice result. We will work with a simple 3 3 real symmetric
matrix with 3 distinct eigenvalues simply to make the notation less cluttered than would
be the case if we leapt straight into the general NN case. We will have 3 eigenvalues ,
and . The corresponding eigenvectors will be u, v and w. Each eigenvector contains
three numbers, so we will write v = [v
1
, v
2
, v
3
]
T
etc. We are free to stretch or shrink each
eigenvector so let us assume that they have been scaled so that each is a unit vector,
i.e. v
T
v = 1 etc. Now lets assemble the three separate eigenvalue equations into one
big matrix equation, like this
A
_
_
u
1
v
1
w
1
u
2
v
2
w
2
u
3
v
3
w
3
_
_
=
_
_
u
1
v
1
w
1
u
2
v
2
w
2
u
3
v
3
w
3
_
_
_
_
0 0
0 0
0 0
_
_
This looks pretty but what can we do with this? Good question. The big trick is that we
can easily (trust me) solve this set of equations for the matrix A. Really? Lets suppose
that the 3 3 matrix to the right of A has an inverse. Then we could solve for A by
multiplying by the inverse from the left, to obtain
A =
_
_
u
1
v
1
w
1
u
2
v
2
w
2
u
3
v
3
w
3
_
_
_
_
0 0
0 0
0 0
_
_
_
_
u
1
v
1
w
1
u
2
v
2
w
2
u
3
v
3
w
3
_
_
1
This is nice, but can we compute the inverse? In fact we already have it, just look
carefully at this equation
_
_
u
1
u
2
u
3
v
1
v
2
v
3
w
1
w
2
w
3
_
_
_
_
u
1
v
1
w
1
u
2
v
2
w
2
u
3
v
3
w
3
_
_
=
_
_
1 0 0
0 1 0
0 0 1
_
_
This is just a simple way of stating that the eigenvectors are orthogonal and of unit
length. This also shows that one matrix is the inverse of the other, that is
_
_
u
1
v
1
w
1
u
2
v
2
w
2
u
3
v
3
w
3
_
_
1
=
_
_
u
1
u
2
u
3
v
1
v
2
v
3
w
1
w
2
w
3
_
_
26-Jul-2013 50
School of Mathematical Sciences Monash University
Now we have our nal result
A =
_
_
u
1
v
1
w
1
u
2
v
2
w
2
u
3
v
3
w
3
_
_
_
_
0 0
0 0
0 0
_
_
_
_
u
1
u
2
u
3
v
1
v
2
v
3
w
1
w
2
w
3
_
_
This shows that any real symmetric 3 3 matrix, with three distinct eigenvalues, can
be re-built from its eigenvalues and eigenvectors. This is not only a neat result it is also
an extremely useful result.
In the following examples we will assume that the matrix A is a real symmetric 3 3
matrix with three distinct eigenvalues.
Example 8.12
Use the above expansion for A to compute A
2
, A
3
, A
4
and so on.
Example 8.13
Use the denition of an eigenvalue to show that A
2
has an eigenvalue
2
, A
3
an eigenvalue

3
and so on. How does this compare with the previous example?
Example 8.14
Suppose that is an eigenvalue, with corresponding eigenvector v, of any square matrix
B. Can you construct an eigenvalue and eigenvector for B
1
(assuming that the inverse
exists)?
8.4 Matrix inverse
The past few examples shows, for our general class of real symmetric 3 3 matrices A,
with three distinct eigenvalues, that the powers of A can be written as
A
n
=
_
_
u
1
v
1
w
1
u
2
v
2
w
2
u
3
v
3
w
3
_
_
_
_

n
0 0
0
n
0
0 0
n
_
_
_
_
u
1
u
2
u
3
v
1
v
2
v
3
w
1
w
2
w
3
_
_
It is easy to see that this is true for any positive integer n. But it also applies (assuming
, and are non-zero) when n is a negative integer. How can we be so sure? We
know that A and A
1
share the same eigenvectors. Good. We also know that if is an
eigenvalue of A then 1/ is an eigenvalue of A
1
. Finally we note that A
1
, like A, is a
real symmetric 3 3 matrix with three (non-zero) distinct eigenvalues. Since we know
all of its eigenvalues and eigenvectors we can use the eigenvalue expansion to write A
1
as
A
1
=
_
_
u
1
v
1
w
1
u
2
v
2
w
2
u
3
v
3
w
3
_
_
_
_

1
0 0
0
1
0
0 0
1
_
_
_
_
u
1
u
2
u
3
v
1
v
2
v
3
w
1
w
2
w
3
_
_
26-Jul-2013 51
School of Mathematical Sciences Monash University
Which is just what we would have got by putting n = 1 in the previous equation.
From here we could compute A
2
= A
1
A
1
, A
3
= A
1
A
2
and so on. In short, we
have proved the above expression for A
n
for any integer n, positive or negative.
The above result (with n = 1) give us yet another way to compute the inverse of A.
Isnt this exciting (and unexpected)?
8.5 The Cayley-Hamilton theorem: Not examinable
What do we know about the three eigenvalues , and ? We know that they are
solutions of the characteristic polynomial
0 = det(A I)
which, after some simple algebra, leads to a polynomial of the form
0 =
3
+ b
1

2
+ b
2

1
+ b
3

0
where b
1
, b
2
and b
3
are some numbers (built from the numbers in A).
Now lets do something un-expected (expect the un-expected). Lets replace the number
with the 3 3 matrix A in the right hand side of the above polynomial. Where we
encounter the powers of A we will use what we have learnt above, that we can use
expansions in powers of the eigenvalues. Thus we have
A
3
+ b
1
A
2
+ b
2
A + b
3
I =
_
_
u
1
v
1
w
1
u
2
v
2
w
2
u
3
v
3
w
3
_
_
_
_

3
0 0
0
3
0
0 0
3
_
_
_
_
u
1
u
2
u
3
v
1
v
2
v
3
w
1
w
2
w
3
_
_
+ b
1
_
_
u
1
v
1
w
1
u
2
v
2
w
2
u
3
v
3
w
3
_
_
_
_

2
0 0
0
2
0
0 0
2
_
_
_
_
u
1
u
2
u
3
v
1
v
2
v
3
w
1
w
2
w
3
_
_
+ b
2
_
_
u
1
v
1
w
1
u
2
v
2
w
2
u
3
v
3
w
3
_
_
_
_

1
0 0
0
1
0
0 0
1
_
_
_
_
u
1
u
2
u
3
v
1
v
2
v
3
w
1
w
2
w
3
_
_
+ b
3
_
_
u
1
v
1
w
1
u
2
v
2
w
2
u
3
v
3
w
3
_
_
_
_

0
0 0
0
0
0
0 0
0
_
_
_
_
u
1
u
2
u
3
v
1
v
2
v
3
w
1
w
2
w
3
_
_
We can tidy this up by collecting the eigenvalue terms into one matrix
A
3
+ b
1
A
2
+ b
2
A + b
3
I =
_
_
u
1
v
1
w
1
u
2
v
2
w
2
u
3
v
3
w
3
_
_
_
_
D
11
0 0
0 D
22
0
0 0 D
33
_
_
_
_
u
1
u
2
u
3
v
1
v
2
v
3
w
1
w
2
w
3
_
_
where
D
11
=
3
+ b
1

2
+ b
2
+ b
3
D
22
=
2
+ b
1

2
+ b
2
+ b
3
D
33
=
3
+ b
1

2
+ b
2
+ b
3
26-Jul-2013 52
School of Mathematical Sciences Monash University
However we know that each eigenvalue is a solution of the polynomial equation
0 =
3
+ b
1

2
+ b
2
+ b
3
which means that D
11
= D
22
= D
33
= 0 and thus the middle matrix is in fact the zero
matrix. Thus we have shown that
0 = A
3
+ b
1
A
2
+ b
2
A + b
3
I
This is an example of the Cayley-Hamilton theorem. It is very much un-expected
(agreed?).
It has been a long road but the journey was fun (yes it was) and it has lead us to a
famous theorem in the theory of matrices, the Cayley-Hamilton theorem. Though we
have demonstrated the theorem for the particular case of real symmetric matrices with
distinct eigenvalues it, the theorem, happens to be true for any square matrix. Proving
that this is so is far from easy but sadly the margins of this textbook are too narrow to
record the proof, you will have to wait until your second year of maths.
The Cayley-Hamilton theorem
Let A be any N N matrix. Then dene the polynomial P() by
P() = det(A I)
where I is the N N identity matrix. Then
0 = P(A)
Note that the eigenvalues of A are the solutions of
0 = P()
26-Jul-2013 53
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
9. Integration
School of Mathematical Sciences Monash University
9.1 Integration : Revision
Computing I =
_
f(x)dx is no dierent from nding the function F(x) such that
dF/dx = f(x).
The function F(x) is called the anti-derivative of f(x). Finding F(x) can be very tricky.
Example 9.1
I =
_
sin x dx
This means nd the function F(x) such that
dF(x)
dx
= sin x
We know this to be F(x) = cos x + C where C is a constant of integration.
Example 9.2
I =
_
sin(3x) dx
For this we use a substitution,
u = 3x , du = 3dx , dx =
du
3
Thus we have
I =
_
sin(3x) dx =
_
(sin u)
_
1
3
du
_
=
1
3
_
sin u du
=
1
3
(cos u) + C
Now we ip back to the variable x,
I =
_
sin(3x) dx =
1
3
cos(3x) + C
Example 9.3
I =
_
x exp(x
2
) dx
Choose a substitution that targets the ugly bit in the integral. Thus put u(x) = x
2
.
Then du = 2xdx and xdx = du/2. This gives us
I =
_
1
2
exp(u) du =
1
2
exp u + C =
1
2
exp(x
2
) + C
26-Jul-2013 55
School of Mathematical Sciences Monash University
9.1.1 Some basic integrals
You must remember the following integrals.
_
exp(x) dx = exp(x) + C
_
cos(x) dx = sin(x) + C
_
sin(x) dx = cos(x) + C
_
x
n
dx =
1
n + 1
x
n+1
, n = 1
_
1
x
dx = log(x) + C
9.1.2 Substitution
If I =
_
f(x) dx looks nasty, try changing the variable of integration. That is put
u = u(x) for some chosen function u(x) (usually inspired by some part of f(x)). Then
we invert the function to nd x = x(u) and substitute into the integral.
I =
_
f(x) dx =
_
f(x(u))
dx
du
du
If we have chosen well, then this second integral will be easy to do.
9.2 Integration by parts
This is a very powerful technique based upon the product rule for derivatives.
Recall that
d(fg)
dx
= g
df
dx
+ f
dg
dx
Now integrate both sides
_
d(fg)
dx
dx =
_
g
df
dx
dx +
_
f
dg
dx
dx
But integration is the inverse of dierentiation, thus we have
fg =
_
g
df
dx
dx +
_
f
dg
dx
dx
26-Jul-2013 56
School of Mathematical Sciences Monash University
which we can re-arrange to
_
f
dg
dx
dx = fg
_
g
df
dx
dx
Thus we have converted one integral into another. The hope is that the second integral
is easier than the rst. This will depend on the choices we make for f and dg/dx.
Example 9.4
I =
_
x exp(x) dx
We have to split the integrand x exp(x) into two pieces, f and dg/dx.
Choose
f(x) = x
df
dx
= 1
dg
dx
= exp(x) g(x) = exp(x)
Then
I =
_
x exp(x) dx = fg
_
g
df
dx
dx
= x exp(x)
_
1 exp(x) dx
= x exp(x) exp(x) + C
Example 9.5
I =
_
x cos(x) dx
Choose
f(x) = x
df
dx
= 1
dg
dx
= cos(x) g(x) = sin(x)
and thus
I =
_
x cos(x) dx = x sin(x)
_
1 sin(x) dx
= x sin(x) + cos(x) + C
26-Jul-2013 57
School of Mathematical Sciences Monash University
Example 9.6
I =
_
x log(x) dx
Choose
f(x) = x
df
dx
= 1
dg
dx
= log(x) g(x) =???
We dont know immediately the anti-derivative for log(x), so we try another split. This
time we choose
f(x) = log(x)
df
dx
=
1
x
dg
dx
= x g(x) =
1
2
x
2
and this leads to
I =
_
x log(x) dx =
1
2
x
2
log(x)
_
1
2
x
2
1
x
dx
=
1
2
x
2
log(x)
1
4
x
2
+ C
Example 9.7
Evaluate
I =
_
log(x)
x
dx
Two students were doing a maths test. The answer to the the rst question
was log(1 + x). A weak student copied the answer from a good student,
but didnt want to make it obvious that he was cheating, so he changed the
answer slightly, to timber(1 + x).
26-Jul-2013 58
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
10. Hyperbolic functions
School of Mathematical Sciences Monash University
10.1 Hyperbolic functions
Do you remember the time when you rst encountered the sine and cosine functions?
That would have been in early secondary school when you were studying trigonometry.
These functions proved very useful when faced with problems to do with triangles. You
may have been surprised when (many years later) you found that those same functions
also proved useful when solving some integration problems. Here is a classic example.
Example 10.1 Integration requiring trigonometric functions
Evaluate the following anti-derivative
I =
_
1

1 x
2
dx
We will use a substitution, x(u) = sin u, as follows
I =
_
1

1 x
2
dx put x = sin u and dx = cos u du
=
_
1
cos u
cos u du
=
_
du
and thus _
1

1 x
2
dx = sin
1
x
where, for simplicity, we have ignored the usual integration constant.
This example was very simple and contained nothing new. But if we had been given the
following integral
I =
_
1

1 + x
2
dx
and continued to use a substitution based on simple sine and cosine functions then
we would nd the game to be rather drawn out. As you can easily verify, the correct
substitution is x(u) = tan u and the integration (ignoring integration constants) leads
to _
1

1 + x
2
dx = log
e
_
x +

1 + x
2
_
26-Jul-2013 60
School of Mathematical Sciences Monash University
Example 10.2
Verify the above integration.
This situation is not all that satisfactory as it involve a series of tedious substitutions
and takes far more work than the rst example. Can we do a better job? Yes, but
it involves a trick where we dene new functions, known as hyperbolic functions, to do
exactly that job.
For the moment we will leave behind the issue of integration and focus on this new class
of functions. Later we will return to our integrals to show how easy the job can be.
10.1.1 Hyperbolic functions
The hyperbolic functions are rather easy to dene. It all begins with this pair of functions
sinh u, known as hyperbolic sine and pronounced either as sinch or shine and cosh u,
known as hyperbolic cosine and pronounced as cosh. They are dened by
sinh u =
1
2
_
e
u
e
u
_
cosh u =
1
2
_
e
u
+ e
u
_
|u| <
These functions bare names similar to sin and cos for the simple reason that they share
properties similar to those of sin and cos (as we will soon see).
The above denitions for sinh and cosh are really all you need to know everything else
about hyperbolic functions follows form these two denitions. Of course it does not hurt
to commit to memory some of the equations we are about to present.
Here are a few elementary properties of sinh and cosh You can easily verify that
cosh
2
u sinh
2
u = 1
d cosh u
du
= sinh u ,
d sinh u
du
= cosh u
Here is a more detailed list of properties (which of course you will verify, by using the
above denitions).
Properties of Hyperbolic functions. Pt.1
cosh
2
x sinh
2
x = 1
cosh(u + v) = cosh ucosh v + sinh usinh v
sinh(u v) = sinh ucosh v sinh v cosh u
2 cosh
2
x = 1 + cosh(2x) , 2 sinh
2
x = 1 + cosh(2x)
d cosh x
dx
= sinh x ,
d sinh x
dx
= cosh x
26-Jul-2013 61
School of Mathematical Sciences Monash University
x
c
o
s
h
,
s
i
n
h
cosh(x)
sinh(x)
3 2 1 0 1 2 3

1
0

5
0
5
1
0
This looks very pretty and reminds us (well it should remind us) of remarkably similar
properties for the sin and cos functions. Now recall the promise we gave earlier, that
these hyperbolic functions would make our life with certain integrals much easier. So let
us return to the integral from earlier in this chapter. Using the same layout and similar
sentences here is how we would complete the integral using our new found friends.
Example 10.3 Integration requiring hyperbolic functions
Evaluate the following anti-derivative
I =
_
1

1 + x
2
dx
We will use a substitution, x(u) = sinh u, as follows
I =
_
1

1 + x
2
dx put x = sinh u and dx = cosh u du
=
_
1
cosh u
cosh u du
=
_
du
and thus _
1

1 + x
2
dx = sinh
1
x
where, for simplicity, we have ignored the usual integration constant.
10.1.2 More hyperbolic functions
You might be wondering if there are hyperbolic equivalents to the familiar trigonometric
functions tan, cotan, sec and cosec. Good question, and yes, indeed there are equivalents
26-Jul-2013 62
School of Mathematical Sciences Monash University
named tanh, cotanh, sech and cosech. The following table provides some basic facts
(which again you should verify).
Properties of Hyperbolic functions. Pt.2
tanh x =
sinh x
cosh x
cotanh x =
cosh x
sinh x
sech x =
1
cosh x
cosech x =
1
sinh x
sech
2
x tanh
2
x = 1
d tanh x
dx
= sech
2
x ,
d cotanh x
dx
= cosech
2
x
10.2 Special functions: not examinable
In the previous examples we conveniently ignored the integration constants. But we
should not be so ippant, instead we should have written
sinh
1
x = C +
_
x
0
1

1 + u
2
du
Note that the integral on the right hand side vanishes when x = 0 and thus C =
sinh
1
(0). The good thing is that we know that sinh(0) = 0 and this fact can be used
to properly determine the integration constant, that is C = 0 and thus we have
sinh
1
x =
_
x
0
1

1 + u
2
du
Now we come to an interesting re-interpretation. We could have begun our discussions
on hyperbolic sine from this very equation. That is, we could use the right hand side
to dene the (inverse) hyperbolic sine. But now you might ask: How do we compute
a number for sinh
1
(0.45)? One method would be to compute an approximation by
estimating the area under the curve. A better method is to evaluate the right hand side
using log
e
(x +

1 + x
2
) as the anti-derivative. Either way it is a bit messy but it does
establish the point, that this integral contains everything we could ever wish to know
about sinh
1
.
What is the point of this discussion? Well it shows how we can turn adversity into
advantage. Where previously we had a dicult integral (not impossible but dicult
none the less) we invented new functions (the hyperbolic functions) that made such
integrals trivial. The same idea can be applied to many many more integrals. For
example, the following integral
erf(x) =
2

_
x
0
e
u
2
du
26-Jul-2013 63
School of Mathematical Sciences Monash University
denes a special function known as the error function. It is used extensively in statistics
and diusion problems (such as the ow of heat). For this integral there is no known
anti-derivative and thus values for erf(x) can only be obtained by some other means
(e.g., the area under the graph).
Euclids fashion tip: When attending Hyperbolic Functions always choose neat and
casual.
26-Jul-2013 64
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
11. Improper integrals
School of Mathematical Sciences Monash University
11.1 Improper Integrals
When a denite integral contains an innity, either in the integrand or on the limits, we
say that we have an improper integral. All other integrals are proper integrals.
Example 11.1
I =
_
1
0
dx

x
, I =
_

0
dx
1 + x
2
I =
_
1
1
dx
x
2
, I =
_
/2
0
tan(x) dx
Because of the innity we must treat the improper integral with care.
11.1.1 A standard strategy
We construct a related proper integral say I() that depends on a parameter . We
choose I() such that we recover the original integral as a limit, say as 0.
Example 11.2
I =
_
1
0
dx

x
For this we construct a related proper integral
I() =
_
1

dx

x
> 0
Since this is a proper integral for > 0 we can evaluate it directly,
I() =
_
1

dx

x
=
_
2

= 2 2

Next we evaluate the limit as 0


lim
0
I() = lim
0
2 2

= 2
As this answer is well dened (i.e. nite and independent of the way the limit is ap-
proached) we are justied in dening this to be the value of the improper integral.
I =
_
1
0
dx

x
= lim
0
_
1

dx

x
= 2
26-Jul-2013 66
School of Mathematical Sciences Monash University
In this case we say we have a convergent improper integral. Had we not got a nite
answer we would say that we had an divergent improper integral.
Example 11.3
I =
_
1
0
dx
x
2
This time we choose
I() =
_
1

dx
x
2
> 0
which we can easily evaluate
I() =
1

1
and thus
lim
0
I() =
This is not nite so this time we say the the improper integral is a divergent improper
integral.
Example 11.4
What do you think of the following calculation? Be warned : the answer is wrong!
I =
_
1
1
dx
x
2
=
_
1
x
_
1
1
= 2
Example 11.5
I =
_
1
1
dx
x
3
This time we have an improper integral because the integrand is singular inside the
region of integration. We create our related proper integral by cutting out the singular
point. Thus we dene two separate proper integrals,
26-Jul-2013 67
School of Mathematical Sciences Monash University
I
1
() =
_

1
dx
x
3
dx > 0
I
2
() =
_
1

dx
x
3
dx > 0
If both I
1
and I
2
converge (i.e. have nite values) we say that I also converges with the
value
I = lim
0
I
1
() + lim
0
I
2
()
But for our case
I
1
() = 1
1

2
lim
0
I
1
() =
I
2
() = 1 +
1

2
lim
0
I
2
() = +
Thus neither I
1
nor I
2
converges and thus I is a divergent improper integral.
This may seem easy (it is) but it does require some care as the next example shows.
Example 11.6
Suppose we chose I
1
and I
2
as before but we set
=

_
1 + 2
2

2

1

2
= 2
Then we would nd that
I
1
() + I
2
() =
_
1

2

1

2
_
= 2
for all > 0 and therefore
lim
0
I
1
+ I
2
= 2
But had we chosen
=
26-Jul-2013 68
School of Mathematical Sciences Monash University
we would have found that
lim
0
I
1
+ I
2
= 0
How can this be? The answer is that in computing I
1
+ I
2
we are eventually trying to
make sense of +. Depending on how we approach the limit we can get any answer
we like for +.
Consequently, when we say that an integral is divergent we mean that either its value
is innity or that it has no single well dened value.
A father who is very much concerned about his sons bad grades in maths
decides to register him at a catholic school. After his rst term there, the son
brings home his report card: Hes getting As in maths. The father is, of
course, pleased, but wants to know: Why are your maths grades suddenly
so good? You know, the son explains, the rst day when I walked into
the classroom, I instantly that this place means business! And why was that
inquired the father. Dad! theyve got a guy hanging on the wall and theyve
nailed him to a plus sign! exclaimed his son.
26-Jul-2013 69
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
12. Comparison test for convergence
School of Mathematical Sciences Monash University
12.1 Comparison Test for Improper Integrals
Example 12.1
I =
_

2
e
x
2
dx
For this integral we would choose
I() =
_

2
e
x
2
dx
and provided that the limit exists, we would write
I = lim

I()
The trouble is we do not have a simple anti-derivative for e
x
2
. The trick here is to look
at a simpler (improper) integral for which we can nd a simple anti-derivative.
Note that
0 < e
x
2
< e
x
for 2 < x <
Now integrate
0 <
_

2
e
x
2
dx <
_

2
e
x
dx
and the last integral on the right is easy to do (thats one reason why we chose e
x
),
0 <
_

2
e
x
2
dx <
_

2
e
x
dx = e
2
e

Our next step is take the limit as


0 < lim

_

2
e
x
2
dx < lim

e
2
e

= e
2
The limit exists and is nite so we have our nal answer that I =
_

2
e
x
2
dx is
convergent.
26-Jul-2013 71
School of Mathematical Sciences Monash University
Example 12.2
I =
_
1
0
e
x
x
dx and I() =
_
1

e
x
x
dx
Again we do not have a simple anti-derivative for e
x
/x so we study a related integral
J =
_
1
0
1
x
dx and J() =
_
1

1
x
dx
For each integral the appropriate limit is 0.
Now we proceed as follows
0 <
1
x
<
e
x
x
for 0 < x < 1
0 <
_
1

1
x
dx <
_
1

e
x
x
dx
0 < log 1 log < I()
Now we take the limit, in this case, 0,
0 < lim
0
(log 1 log ) < lim
0
I()
0 < < lim
0
I()
Thus we conclude that I =
_
1
0
e
x
/x dx is divergent.
Example 12.3
I =
_
1
0
e
x
x
dx and I() =
_
1

e
x
x
dx
Suppose (mistakenly) we thought that this integral converged. We might set out to
prove this by starting with
0 <
e
x
x
<
3
x
for 0 < x < 1
then we would leap into the now familiar steps,
26-Jul-2013 72
School of Mathematical Sciences Monash University
0 <
_
1

e
x
x
dx <
_
1

3
x
dx = 3(log 1 log )
0 < lim
0
_
1

e
x
x
dx < lim
0
3(log 1 log ) =
0 < lim
0
I() <
This last line tells us nothing. Though we set out to prove convergence we actually
proved nothing. Thus either we were wrong in supposing that the integral converged or
we made a bad choice for the test function 3/x. We know from the previous example
that in fact this integral is divergent.
12.2 The General Strategy
Suppose we have
_

0
f(x) dx with f(x) > 0
Then we have two cases to consider.
Test for convergence If you can nd c(x) such that
(1) 0 < f(x) < c(x) and
(2) lim

0
c(x) dx is nite.
then I is convergent.
26-Jul-2013 73
School of Mathematical Sciences Monash University
Test for divergence If you can nd d(x) such that
(1) 0 < d(x) < f(x) and
(2) lim

0
d(x) dx = .
then I is divergent.
We generally try to choose the test function (c(x) or d(x)) so that it has a simple
anti-derivative.
A strategy similar to the above would apply for integrals like I =
_
1
0
f(x) dx.
Example 12.4
Re-write the above strategy for the case I =
_
1
0
f(x) dx.
A mathematician named Klein
Thought the Mobius band was divine
Said he: If you glue
The edges of two
Youll get a weird bottles like mine.
26-Jul-2013 74
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
13. Introduction to sequences and series.
School of Mathematical Sciences Monash University
13.1 Denitions
Sequence. A set of numbers such as
1,
1
2
,
1
3
,
1
4
,
1
5
,
1,
1
2
,
1
3
,
1
4
,
1
5
,
1,
1
4
,
1
9
,
1
16
,
1
25
,
Each term in the sequence is often denoted by a subscripted symbol,
a
n
=
1
n+1
, n = 0, 1, 2, 123
b
n
=
(1)
n
n+1
, n = 0, 1, 2, 666
c
n
=
1
(n+1)
2
, n = 0, 1, 2,
The rst two sequences have a nite number of terms while the the last sequence
is innitely long.
Series. The sum of terms that dene a sequence,
1 +
1
2
+
1
3
+
1
4
+
1
5
+ +
1
123
1
1
2
+
1
3

1
4
+
1
5

1
666
1 +
1
4
+
1
9
+
1
16
+
1
25
+ +
1

The rst two series are nite series while the last is an innite series.
13.1.1 Notation
The terms in a sequence are normally counted from zero, that is we have a
0
, a
1
, a
2
, .
For an innite series we usually include just the rst three terms, followed by three
dots to indicate that there are more terms, then the generic term and nally three
more dots to remind us that its an innite series. Thus the last example above
would normally be written as
1 +
1
4
+
1
9
+ +
1
n
2
+
13.1.2 Partial sums
Given a sequence dened by a
n
we can form a new sequence by adding together the
successive a
n
, that is
S
n
= a
0
+ a
1
+ a
2
+ + a
n
26-Jul-2013 76
School of Mathematical Sciences Monash University
Each S
n
is a nite sum of numbers. The really interesting question is what happens to
S
n
as n ? For example, you might think that the innite series
1 +
1
2
+
1
3
+ +
1
n

might be nite because the terms in the tail go to zero but youd be wrong, this series
has no nite value (as we shall see in a later example). On the other hand, the innite
series
1
1
2
+
1
3
+
(1)
n+1
n

does have a well dened nite value. The general approach to understanding which case
we have is to examine the limit of the sequence of partial sums S
n
as n . This we
shall study in detail soon, but rst well play with some preliminary examples.
13.1.3 Arithmetic series
This is about as simple as it gets, each new term in the series diers from the previous
term by a constant number d. Thus if the rst term is a
0
= a then we have
a
n
= a
n1
+ d = a + nd
S
n
= a
0
+ a
1
+ a
2
+ + a
n
Its easy to show that
S
n
= (n + 1)a +
1
2
n(n + 1)d
Example 13.1
Verify the above formula for S
n
.
13.1.4 Fibonacci sequence
In this sequence each new number is generated as the sum of the two previous numbers,
for example, 0, 1, 1, 2, 3, 5, 8, 13, 21 . The general term in the Fibonacci sequence is
often written as F
n
, with
F
n
= F
n1
+ F
n2
26-Jul-2013 77
School of Mathematical Sciences Monash University
Example 13.2
Construct the new sequence G
n
= F
n
/F
n1
. Show that
lim
n
G
n
= (1 +

5)/2
13.1.5 Geometric series
This is similar to the arithmetic series with the exception that each new term is a
multiple of the previous term. Thus we have
a
n
= sa
n1
= a
0
s
n
S
n
= a
0
+ a
1
+ a
2
+ + a
n
= a
0
(1 + s + s
2
+ s
3
+ + s
n
)
For this series we have
S
n
=
_

_
(n + 1)a
0
: s = 1
(1 s
n+1
) a
0
1 s
: s = 1
The parameters a
0
and s are known as the initial value and common ratio respectively.
Example 13.3
Prove the above formula for S
n
.
Example 13.4
Two trains 200 km apart are moving toward each other; each one is going at a speed
of 50 km/hr. A y starting on the front of one of them ies back and forth between
them at a rate of 75 km/hr (its fast!). It does this until the trains collide. What is the
total distance the y has own? (No animals were harmed in this example, its just a
hypothetical example!)
Example 13.5
The previous problem can be solved using an innite geometric series. Is there another,
quicker, way?
13.1.6 Compound Interest
Suppose you have a very generous (or silly) bank manager. Suppose he/she oers you
10% compound interest per year on your savings. You start with $ 100 and then you sit
26-Jul-2013 78
School of Mathematical Sciences Monash University
back and do nothing (other than to plough the interest earned back into your account
and watch your savings grow).
How much money will you have after 10 years?
Let S
n
be the savings at the end of year n. Then we have
S
0
= $100
S
1
= S
0
+ 0.10 S
0
= $110
S
2
= S
1
+ 0.10 S
1
= $121
S
3
= S
2
+ 0.10 S
2
= $133
.
.
.
.
.
.
S
10
= S
9
+ 0.10 S
9
= $260
Not a bad return for no work in ten years (pity interest rates for savings are not at 10%).
A mathematician organizes a rae in which the prize is an innite amount of
money paid over an innite amount of time. Of course, with the promise of
such a prize, his tickets sell like hot cake. When the winning ticket is drawn,
and the jubilant winner comes to claim his prize, the mathematician explains
the mode of payment: 1 dollar now, 1/2 dollar next week, 1/3 dollar the
week after that...
26-Jul-2013 79
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
14. Convergence of series.
School of Mathematical Sciences Monash University
14.1 Innite series
The main issue with most innite series is whether or not the series converges. Of
secondary importance is what the sum of that series might be, assuming it to be a
convergent series.
14.1.1 Convergence and divergence
Let S = a
0
+ a
1
+ a
2
+ + a
n
+ =

k=0
a
k
be an innite series. Let S
n
=
a
0
+ a
1
+ a
2
+ + a
n
be the partial sum for S, then
Convergence. The innite series converges when lim
n
S
n
exists and is nite.
Divergence. In all other cases we say that the series diverges.
14.2 Tests for convergence
Though the integral test is very powerful it is not the only test that we can apply when
testing an innite series for convergence. Here are some such tests.
In all cases, we are interested in the behaviour of a
n
and S
n
for large values of n. It does
not really matter what happens for small n, its the behaviour of tail that counts.
In all of the following tests we are trying to establish whether the innite series S =

0
a
n
converges or diverges.
14.2.1 Zero tail?
This is as simple as it gets. If the a
n
do not vanish as n then the innite series
diverges. This should be obvious if the tail does not diminish to zero then we must be
adding on a nite term at the end of the series and hence the series can not settle down
to one xed number.
This condition, that a
n
0 as n for the series to converge, is known as a necessary
condition.
Note that this condition tells us nothing about the convergence of the series when a
n
0
as n .
14.2.2 The Comparison test
Suppose we have two other innite series, a convergent series C =

n=0
c
n
and a
divergent series D =

n=0
d
n
.
Then we will have
26-Jul-2013 81
School of Mathematical Sciences Monash University
Convergence : When 0 < a
n
< c
n
for all n > m
Divergence : When 0 < d
n
< a
n
for all n > m
where m is some (possibly large) number.
Example 14.1
Use the comparison test to show that the innite series
S =

n=0
1
n
2
+ n + 1
is convergent.
Example 14.2
Repeat the previous example, but this time use the integral test.
26-Jul-2013 82
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
15. Integral and ratio tests.
School of Mathematical Sciences Monash University
Example 15.1
It was claimed earlier that
S = 1 +
1
2
+
1
3
+ +
1
n
+ =

k=1
1
k
diverges. How can we be so sure?
Here is a table of the partial sums
n 10 100 1000 10000 100000 1000000
S
n
2.92897 5.18738 7.48547 9.78761 12.0901 14.3927
This seems to suggest that S
n
increases by a (nearly) constant amount every time we
multiply the number of terms by 10. This is not a proof that the series diverges, but it
is a strong indication!
How might we estimate S
n
? There is a trick (yes, another trick) in which we use
a specially constructed set of rectangles to estimate S
n
by computing the total area
contained in those rectangles.
x
y
(
x
)
Over estimate
Under estimate
y(x) = 1/x
0 2 4 6 8 10 12
0
.
2
0
.
4
0
.
6
0
.
8
1
.
0
For the series 1 + 1/2 + 1/3 + 1/n we can construct two sets of rectangles, one
set lies below the curve y(x) = 1/x, the other set lies above (as in the diagram). The
crucial point is that as each rectangle has unit width and height equal to 1/x its area
matches exactly the value of a term in the series. Thus we have the following
26-Jul-2013 84
School of Mathematical Sciences Monash University
Below rectangles : B(n) = 1 +
1
2
+
1
3
+ +
1
n1
= S
n1
Above rectangles : A(n) =
1
2
+
1
3
+ +
1
n
= S
n
1
Clearly we also have
B(n) <
_
n
1
1
x
dx < A(n)
but
_
n
1
1
x
dx = log(n)
and thus
S
n
1 < log(n) < S
n1
Now we can take the limit as n ,
lim
n
S
n
1 < lim
n
log(n) < lim
n
S
n1
lim
n
S
n
1 < < lim
n
S
n1
The left hand side of this inequality doesnt tell us much but the right hand side shows
clearly that lim
n
S
n1
is innity. That is, the partial sums S
n
do not converge and
the original series is divergent.
Note that the series
S = 1 +
1
2
+
1
3
+ +
1
n
+ =

k=1
1
k
is known as the Harmonic series.
15.1 The Integral Test
The previous example introduced a very powerful method for deciding if a series con-
verges or diverges. The general approach is as follows.
Suppose we have an innite series
S =

k=0
a
k
26-Jul-2013 85
School of Mathematical Sciences Monash University
and that the terms can be computed from a smooth function f(x) with a
k
= f(k) for
k = 0, 1, 2 . Then
S
n
a
0
<
_
n
0
f(x) dx < S
n1
We then take the limit as n ,
lim
n
S
n
a
0
<
_

0
f(x) dx < lim
n
S
n1
If we can evaluate the improper integral (or at least decide if it converges) then we will
be able to make a decision about the convergence of the innite series.
Convergence. If the improper integral converges then the innite series con-
verges.
Divergence. If the improper integral diverges then so too does the innite series.
Example 15.2
Show that
S =

k=1
1
k
2
is a convergent innite series.
Example 15.3
For what values of p will the innite series
S =

k=1
k
p
converge?
15.2 The Ratio test
This test rst asks you to compute the limit
L = lim
n
a
n+1
a
n
then we have
26-Jul-2013 86
School of Mathematical Sciences Monash University
Convergence : When L < 1
Divergence : When L > 1
Indeterminate : When L = 1
Example 15.4
Show, using the ratio test, that the geometric series S =

n=0
s
n
is convergent when
0 < s < 1.
Example 15.5
Show that
S =

n=0
2
n
n
2
+ 1
is a divergent series.
Example 15.6
What does the ratio test tell you about the Harmonic series?
26-Jul-2013 87
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
16. Comparison test, alternating series.
School of Mathematical Sciences Monash University
16.1 Alternating series
So far we have only considered innite series in which all of the terms were positive. But
there are many innite series in which the terms alternate from positive to negative.
Here are two examples
S = 1
1
2
+
1
3

1
4
=

n=0
(1)
n
n + 1
S

= 1
1
3!
+
1
5!

1
7!
=

n=0
(1)
n
(2n + 1)!
These are known as alternating series. There is one very simple test for the convergence
of an alternating series.
If |a
n
| 0 as n then we have
Convergence : When |a
n+1
| < |a
n
| for all n > m
Divergence : In all other cases
where m is some (possibly large) number.
Example 16.1
Use the alternating series test to establish the convergence or otherwise of
S = 1
1
2
+
1
3

1
4
=

n=0
(1)
n
n + 1
Example 16.2
Show that
S = 1
1
3!
+
1
5!

1
7!
=

n=0
(1)
n
(2n + 1)!
is convergent.
Example 16.3
Draw a graph of S
n
versus n for a typical alternating series. Study this graph and
explain why the innite series is convergent when |a
n+1
| < |a
n
|.
26-Jul-2013 89
School of Mathematical Sciences Monash University
16.2 Non-positive innite series
In general, an innite series may contain any mix of positive an negative terms. We
have looked at two classes, one where all the terms are positive and another where the
terms alternate in sign.
What should we do with a general series? Given the general series
S =

n=0
a
n
we construct a related series by taking the absolute value of each term,
S

n=0
|a
n
|
The mix of positive and negative terms in S actually works in favour of convergence for
S. Thus if we can show that S

is convergent then we will also have shown that S is


convergent. On the other if we nd that S

is divergent then we can not say anything


about S.
Convergence : If S

converges, then so to does S


Indeterminate : In all other cases
This type of convergence is known as absolute convergence. If a series is not absolutely
convergent we say that it is conditionally convergent.
Example 16.4
Show that S

< S < S

and hence prove the previous assertion, that S converges


whenever S

converges.
Example 16.5
Show that
S = 1
1
2
2
+
1
3
2

1
4
2
=

n=0
(1)
n
(n + 1)
2
converges.
16.3 Re-ordering an innite series
There is one very useful consequence of absolute convergence.
If a series is absolutely convergent then you may re-order the terms in the innite series
in whatever way you like, the series will converge to the same value every time.
26-Jul-2013 90
School of Mathematical Sciences Monash University
Example 16.6
We know that the series
S = 1
1
2
+
1
3

1
4
=

n=0
(1)
n
n + 1
is convergent. But suppose we grouped all the positive terms into one series and the all
the negative into another,
S = (1 +
1
3
+
1
5
+
1
7
+ )
(
1
2
+
1
4
+
1
6
+ )
= S
+
S

The two series S


+
and S

can be shown (by comparison with the Harmonic series) to


diverge. Thus we would have
S =
and the right hand side is meaningless it can be made to have any value you choose
(by taking suitable limits for the partial sums for S
+
and S

).
The moral is do not re-order a conditionally convergent series!
26-Jul-2013 91
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
17. Power series
School of Mathematical Sciences Monash University
17.1 Simple power series
Here are some typical examples of what are known as power series
f(x) = 1 + x + x
2
+ x
3
+ + x
n
+
g(x) = 1 + x +
x
2
2!
+
x
3
3!
+ +
x
n
n!
+
h(x) = 1
x
2
2!
+
x
4
4!
+ (1)
n
x
2n
(2n)!
+
Each power series is a function of one variable, in this case x and so they are also referred
to as a power series in x.
We might like to ask
For what values of x does the series converge?
If the series converges, what value does it converge to?
The rst question is a simple extension of the ideas we developed in the previous lectures
with the one exception that the convergence of the series may now depend upon the
choice of x.
The second question is generally much harder to answer. We will nd, in the next
lecture, that it is easier to start with a known function and to then build a power series
that has the same values as the function (for values of x for which the power series
converges). By this method (Taylor series) we will see that the three power series above
are representations of the functions f(x) = 1/(1 x), g(x) = e
x
and h(x) = cos(x).
17.2 The general power series
A power series in x around the point x = a is always of the form
a
0
+ a
1
(x a) + a
2
(x a)
2
+ + a
n
(x a)
n
+ =

n=0
a
n
(x a)
n
The point x = a is often said to the be point around which the power series is based.
26-Jul-2013 93
School of Mathematical Sciences Monash University
17.3 Examples of Power Series
In a previous lecture it was claimed that
1
1 x
= 1 + x + x
2
+ x
3
+ + x
n
+
e
x
= 1 + x +
x
2
2!
+
x
3
3!
+ +
x
n
n!
+
cos(x) = 1
x
2
2!
+
x
4
4!
+ (1)
n
x
2n
(2n)!
+
Our game today is to develop a method by which such power series can be constructed.
17.4 Maclaurin Series
Suppose we have a function f(x) and suppose we wish to re-express it as a power series.
That is, we ask if it is possible to nd the coecients a
n
such that
f(x) = a
0
+ a
1
x + a
2
x
2
+ + a
n
x
n
+ =

n=0
a
n
x
n
is valid (for values of x for which the series converges).
Lets just suppose that such an expansion is possible. How might we compute the a
n
?
There is a very neat trick which we will use. Note that if we evaluate both sides of the
equation at x = 0 we get
f(0) = a
0
Thats the rst step. Now for a
1
we rst dierentiate both sides of the equation for f(x),
then put x = 0. The result is
df
dx

x=0
= a
1
And we follow the same steps for all subsequent a
n
. Here is summary of the rst 4 steps.
f(x) = a
0
+ a
1
x + a
2
x
2
+ a
3
x
3
+ f(0) = a
0
f

(x) = a
1
+ 2a
2
x + 3a
3
x
2
+ f

(0) = a
1
f

(x) = 2a
2
+ 6a
3
x + f

(0) = 2a
2
f

(x) = 6a
3
+ f

(0) = 6a
3
26-Jul-2013 94
School of Mathematical Sciences Monash University
A power series developed in this way is known as a Maclaurin Series Here is a general
formula for computing a Maclaurin series.
Maclaurin Series
Let f(x) be an innitely dierentiable function at x = 0. Then
f(x) = a
0
+ a
1
x + a
2
x
2
+ a
3
x
3
+ + a
n
x
n
+
with
a
n
=
1
n!
d
n
f
dx
n

x=0
Example 17.1
Compute the Maclaurin series for log(1 + x).
Example 17.2
Compute the Maclaurin series for sin(x).
17.5 Taylor Series
For a Maclaurin series we are required to compute the function and all its derivatives at
x = 0. But many functions are singular at x = 0 so what should we do in such cases?
Simple choose a dierent point around which to build the power series. Recall that
the general power series for f(x) is of the form
f(x) = a
0
+ a
1
(x a) + a
2
(x a)
2
+ + a
n
(x a)
n
+ =

n=0
a
n
x
n
We can compute the a
n
much as we did in the Maclaurin series with the one exception
that now we evaluate the function and its derivatives at x = a.
Taylor Series
Let f(x) be an innitely dierentiable function at x = a. Then
f(x) = a
0
+ a
1
(x a) + a
2
(x a)
2
+ + a
n
(x a)
n
+
with
a
n
=
1
n!
d
n
f
dx
n

x=a
26-Jul-2013 95
School of Mathematical Sciences Monash University
Example 17.3
Compute the Taylor series for log x around x = 1.
Example 17.4
Compute the Taylor series for sin(x) around x = /2.
17.6 Uniqueness
Is it possible to have two dierent power series for the one function? That is, is it
possible to have
f(x) = a
0
+ a
1
(x a) + a
2
(x a)
2
+ + a
n
(x a)
n
+
and
f(x) = b
0
+ b
1
(x a) + b
2
(x a)
2
+ + b
n
(x a)
n
+
where the a
n
and b
n
are dierent?
The simple answer is no. The coecients of a Taylor series are unique.
What is the use of this fact? It means that regardless of how we happen to compute a
power series we will always obtain the same results.
Example 17.5
Use the Taylor series
e
x
= 1 + x +
x
2
2!
+
x
3
3!
+ +
x
n
n!
+
to compute a power series for e
x
. Compare your result with the Taylor series for e
x
.
Example 17.6
Show how the Taylor series for 1/(1 x) can be used to obtain the Taylor series for
1/(1 x)
2
.
If the proof of a theorem is not immediately apparent, it may be because you are trying
the wrong approach. Below are some eective methods of proof that may aim you in
the right direction.
26-Jul-2013 96
School of Mathematical Sciences Monash University
Methods of Proof in Mathematics
Proof by Example
Obviousness The proof is so clear that it need not be mentioned.
General Agreement All in Favour?...
Imagination Well, lets pretend its true.
Convenience It would be very nice if it were true, so ...
Necessity It had better be true or the whole structure of mathematics would
crumble to the ground.
Plausibility It sounds good so it must be true.
Intimidation Dont be stupid, of course its true.
Lack of Time Because of the time constraint, Ill leave the proof to you.
Postponement The proof for this is so long and arduous, so it is given in the
appendix.
Accident Hey, what have we here?
Insignicance Who really cares anyway?
Mumbo-Jumbo For any > 0 there exists a > 0 such that |f(x) L| <
whenever |x a| < .
Profanity (example omitted)
Lost Reference I know I saw this somewhere ...
Calculus This proof requires calculus, so well skip it.
Lack of Interest Does anyone really want to see this?
Illegibility
Stubbornness I dont care what you say! It is true!
Hasty Generalization Well, it works for 17, so it works for all reals.
Poor Analogy Well, its just like ...
Intuition I just have this gut feeling ...
Vigorous Assertion And I REALLY MEAN THAT!
Divine Intervention Then a miracle occurs ...
26-Jul-2013 97
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
18. Radius of convergence
School of Mathematical Sciences Monash University
18.1 Radius of convergence
If a series converges only for x in the interval |xa| < R, then the radius of convergence
is dened to be R.
Note that it is possible to have R = 0 and even R = .
Example 18.1 : Finite radius of convergence
Consider the power series
f(x) = 1 + x + x
2
+ x
3
+ + x
n
+ =

n=0
x
n
This is the geometric series with common ratio x. We already know that this series
converges when |x| < 1 and thus its radius of convergence is 1.
Example 18.2
Use the ratio test to conrm the previous claim.
Example 18.3
Does the series converge for x = 1? Does it converge for x = 1? (These are minor
dot-the-i-cross-the-t type questions).
Example 18.4 : Innite radius of convergence
Find the radius of convergence for the series
g(x) = 1 + x +
x
2
2!
+
x
3
3!
+ +
x
n
n!
+ =

n=0
x
n
n!
Example 18.5 : Zero radius of convergence
Find the the radius of convergence for the series
Q(x) = 1 + x + 2!x
2
+ 3!x
3
+ + n!x
n
+ =

n=0
n!x
n
18.2 Computing the Radius of Convergence
To compute the radius of convergence R for a power series of the form

0
a
n
(x a)
n
you can take either of two approaches
26-Jul-2013 99
School of Mathematical Sciences Monash University
Direct assault. Use the terms in the power series to dene a new series b
n
=
a
n
(x a)
n
. Then determine the convergence of

n=0
b
n
using any one of the
tests from previous lectures. Note that in this approach the b
n
must be treated as
functions of x.
Use a theorem. If you can compute lim
n
|a
n+1
|/|a
n
| then this limit will be
1/R. Note that this approach does not involve x.
Be careful not to confuse the two approaches!
Example 18.6
Find the radius of convergence for the series
f(x) =

n=0
(1)
n+1
x
n
n
18.3 Some theorems
Each of the following applies for x inside the interval of convergence.
Absolute convergence. If the power series

n=0
|a
n
(x a)
n
| converges then
so to does

n=0
a
n
(x a)
n
.
Term by term dierentiation. A convergent power series may be dierentiated
term by term and it retains the same radius of convergence.
Term by term integration. A convergent power series may be integrated term
by term and it retains the same radius of convergence.
26-Jul-2013 100
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
19. Function Approximation using Taylor Series
School of Mathematical Sciences Monash University
19.1 Motivation
We know that many functions can be written as a Taylor series, including, for example
1
1 x
= 1 + x + x
2
+ x
3
+ + x
n
+
e
x
= 1 + x +
x
2
2!
+
x
3
3!
+ +
x
n
n!
+
cos(x) = 1
x
2
2!
+
x
4
4!
+ (1)
n
x
2n
(2n)!
+
sin(x) = x
x
3
3!
+
x
5
5!
+ (1)
n
x
2n+1
(2n + 1)!
+
Part of our reason for writing functions in this form was that it would allow us to
compute values for the functions (given a value for x).
But each such series is an innite series and so it may take a while to compute every
term! What do we do? Clearly we have to use a nite series. Our plan then is to
truncate the innite series at some point hoping that the terms we leave o contribute
very little to the overall sum.
19.2 Taylor polynomials
Consider the typical Taylor series around x = 0
f(x) = a
0
+ a
1
x + a
2
x
2
+ + a
n
x
n
+ =

k=0
a
k
x
k
We can approximate the innite series by its partial sums. Thus if we dene the Taylor
polynomial by
P
n
(x) = a
0
+ a
1
x + a
2
x
2
+ + a
n
x
n
=
k=n

k=0
a
n
x
n
we can expect each P
n
(x) to be an approximation to f(x) (and only for values of x for
which the innite series converges).
The only question that we really need to ask is How good is the approximation? Here
are some examples.
26-Jul-2013 102
School of Mathematical Sciences Monash University
Example 19.1 : Taylor polynomials for cos(x)
The rst four (distinct) Taylor polynomials for cos(x) are
P
0
(x) = 1
P
2
(x) = 1
x
2
2!
P
4
(x) = 1
x
2
2!
+
x
4
4!
P
6
(x) = 1
x
2
2!
+
x
4
4!

x
6
6!
and this is what they look like
x
y
cos(x)
P
0
(x)
P
2
(x)
P
4
(x)
P
6
(x)
6 4 2 0 2 4 6

1
.
0

0
.
5
0
.
0
0
.
5
1
.
0
Example 19.2
Why were the other Taylor polynomials P
1
, P
3
, P
5
not listed?
26-Jul-2013 103
School of Mathematical Sciences Monash University
Example 19.3
Using the above Taylor polynomials, estimate cos(0.1).
We observe that
All of the approximations are close to cos(x) for x close to zero.
The worst approximation is P
0
(x).
The best approximation is P
6
(x)
We now ask two interesting questions
How would we estimate cos(x) using a Taylor series for x = 12 given that P
0
(x), P
6
(x)
appear to be very poor approximations near x = 12?
Can we estimate the size of the error in approximations?
For the rst question we have two lines of attack. Since we know that cos(x) is periodic
we can use
cos(12) = cos(12 2) = cos(12 4) =
As 12 4 is small we can expect that P
3
(12 4) will be a good approximation to
cos(12).
In general we do not have the luxury of having a periodic function and thus we need an
alternative method.
The second approach is to replace the current Taylor polynomials with a new set built
around x = 4. Thus we put a = 4 in a our general formula for the Taylor series
leading to
P
0
(x) = 1
P
2
(x) = 1
(x 4)
2
2!
P
4
(x) = 1
(x 4)
2
2!
+
(x 4)
4
4!
P
6
(x) = 1
(x 4)
2
2!
+
(x 4)
4
4!

(x 4)
6
6!
26-Jul-2013 104
School of Mathematical Sciences Monash University
x
y
5 0 5 10 15 20

1
.
0

0
.
5
0
.
0
0
.
5
1
.
0
So the lesson is this : Build the Taylor polynomials in the region where you wish to
approximate the function.
19.3 Accuracy
Its all well and good to say that for some values of x the Taylor polynomials yield better
approximations than for other values. It would be far better if we could quantify the
size of the error and identify what parameters eect the quality of the approximation.
This is not easy to do precisely but we can get a feel for what the answers should be.
Let P
n
(x) be a Taylor polynomial around x = a for f(x). Then we have
f(x) =

k=0
a
k
(x a)
k
P
n
(x) =
n

k=0
a
k
(x a)
k
and thus the error in the approximation is
f(x) P
n
(x) = a
n+1
(x a)
n+1
+ a
n+2
(x a)
n+2
+ a
n+3
(x a)
n+3
+
26-Jul-2013 105
School of Mathematical Sciences Monash University
How do we estimate the right hand side? The usual trick is to assume that each each
term is much smaller than its predecessor and thus the right hand side is dominated by
the rst non-zero term.
Thus we often take
f(x) P
n
(x) a
n+1
(x a)
n+1
=
f
(n+1)
(a)
(n + 1)!
(x a)
n+1
where f
(n+1)
(a) is the nth derivative of f(x) at x = a.
This is still a very loose mathematical argument. We have simply ignored all the remain-
ing terms. A better estimate, but harder to justify, is given by the following theorem
Error bound for Taylor polynomials
If P
n
(x) is the Taylor polynomial of degree n for the function f(x) in the interval
|x a| < R then the error E
n
is bounded by
E
n
= |f(x) P
n
(x)| <
M
(n + 1)!
R
n+1
where M is the maximum value of |f
(n+1)
(x)| in the interval |x a| < R.
The upshot of this is that we expect
The error to be zero for polynomials of degree n or less, and
The error, for a xed x, to vary as (x a)
n+1
for varying choices of n.
Example 19.4 : Using the error estimate
Use the error bound to estimate how large the error might be in approximating sin(x)
by P
3
(x) = x x
3
/3! for x in the interval 1 < x < 1.
In this case we have R = 1, a = 0 and n = 3. For M we recall that every derivative of
sin is either sin or cos and thus we can take M = 1. Thus we have
E
3
= | sin(x) P
3
(x)| <
M
(n + 1)!
R
n+1
=
1
4!
1
4
=
1
24
Example 19.5
Use the error bound to estimate how large the error might be in approximating log(1+x)
by P
3
(x) for x in the interval 0 < x < 2. Use a = 1 when computing P
3
.
26-Jul-2013 106
School of Mathematical Sciences Monash University
19.4 Using Taylor series to calculate limits
In your many and varied journeys in the world of mathematics you may have found
statements like
1 = lim
x0
sin(x)
x
and
1 = lim
x1
log(x)
1 x
and you may have been inclined to wonder how such statements can be proved (you do
like to know these things dont you?). Our job in this section is develop a systematic
method by which such hairy computations can be done with modest eort. But rst
a clear warning the following computations apply only to the troublesome
indeterminate form 0/0 (though it is possible to adapt our methods to cases such as
/, well come back to that later). If the calculation that troubles you is not of the
form 0/0 then the following methods will give the wrong answer. Be very careful.
The functions in both of the above examples are of the form f(x)/g(x). Our road to
freedom (from the gloomy prison of 0/0) is to expand f(x) and g(x) as a Taylor series
around the point in question. The limits are then easy to apply. Lets see this in action
for the rst example. Here we have
f(x) = sin x = x
1
3!
x
3
+
1
5!
x
5
+
g(x) = x
In this case the Taylor series for g(x) was rather easy but that isnt always the case.
Thus we have
f(x)
g(x)
=
x
1
3!
x
3
+
1
5!
x
5
+
x
= 1
1
3!
x
2
+
1
5!
x
4
+
and this can be substituted into our expression for the limit,
lim
x0
sin(x)
x
= lim
x0
1
1
3!
x
2
+
1
5!
x
4
+
= 1
For our second example we must employ a Taylor series around x = 1. Thus we have
f(x) = log(x) = (x 1)
1
2
(x 1)
2
+
1
3
(x 1)
3
+
g(x) = 1 x = (x 1)
1
26-Jul-2013 107
School of Mathematical Sciences Monash University
and so
lim
x1
log(x)
1 x
= lim
x1
(x 1)
1
2
(x 1)
2
+
1
3
(x 1)
3
+
(x 1)
= lim
x1
1 +
1
2
(x 1)
1
3
(x 1)
2
+
= 1
This is not all that hard, is it? Here is a slightly trickier example,
lim
x0
1 cos(x)
sin(x
2
)
=?
Once again we build the appropriate Taylor series (in this case around x = 0),
f(x) = 1 cos(x) =
1
2!
x
2

1
4!
x
4
+
1
6!
x
6
+
g(x) = sin(x
2
) = x
2

1
3!
x
6
+
1
5!
x
8
+
and so
lim
x0
1 cos(x)
sin(x
2
)
= lim
x0
1
2!
x
2

1
4!
x
4
+
1
6!
x
6
+
x
2

1
3!
x
6
+
1
5!
x
8
+
= lim
x0
1
2!

1
4!
x
2
+
1
6!
x
4
+
=
1
2
By now the picture should be clear a suitable pair of Taylor series can make short
work of a troublesome 0/0 arising from expressions of the form f(x)/g(x).
19.5 lHopitals rule.
Though the above method works very well it can be a bit tedious. You may have noticed
that our nal answers depended only on the leading terms in the Taylor series and yet we
calculated the whole of the Taylor series. This looks like an un-necessary extra burden.
Can we achieve the same result but with less eort? Most certainly, and here is how we
do it.
26-Jul-2013 108
School of Mathematical Sciences Monash University
lHopitals rule for the form 0/0
If lim
xa
f(x) = 0 and lim
xa
g(x) = 0 then
lim
xa
f(x)
g(x)
= lim
xa
f

(x)
g

(x)
provided the limit exists. This rule can be applied recursively whenever the right
hand side leads to 0/0.
Here is an outline of the proof. We start by writing out the Taylor series for f(x) and
g(x) around x = a (while noting that f(a) = g(a) = 0)
f(x) = f

(a)(x a) +
1
2!
f

(a)(x a)
2
+
1
3!
f

(a)(x a)
3
+
g(x) = g

(a)(x a) +
1
2!
g

(a)(x a)
2
+
1
3!
g

(a)(x a)
3
+
then
f(x)
g(x)
=
f

(a)(x a) +
1
2!
f

(a)(x a)
2
+
1
3!
f

(a)(x a)
3
+
g

(a)(x a) +
1
2!
g

(a)(x a)
2
+
1
3!
g

(a)(x a)
3
+
=
f

(a) +
1
2!
f

(a)(x a) +
1
3!
f

(a)(x a)
2
+
g

(a) +
1
2!
g

(a)(x a) +
1
3!
g

(a)(x a)
2
+
If we assume that g

(a) = 0 then it follows that


lim
xa
f(x)
g(x)
=
f

(a)
g

(a)
This is not exactly lHopitals rule but it gives you an idea of how it was constructed.
With a little more care you can extend this argument to recover the full statement in
lHopitals rule (you need to consider cases where g

(a) = 0).
Example 19.6
Use lHoptals rule to verify the limits that we computed in the three detailed examples
given in the previous section.
We mentioned earlier that the tricks of this section could not only help us make sense of
expressions like 0/0 but also for expressions like /. Without going into the proofs
we will just state the variation of lHopitals rule for cases such as this just do it! Yes,
you can apply lHopitals rule in the same manner as before. Here it is
26-Jul-2013 109
School of Mathematical Sciences Monash University
lHopitals rule for the form /
If lim
xa
f(x) = and lim
xa
g(x) = then
lim
xa
f(x)
g(x)
= lim
xa
f

(x)
g

(x)
provided the limit exists. This rule can be applied recursively whenever the right
hand side leads to /.
Q. Why did the chicken cross the Mobius strip?
A. To get to the same side.
26-Jul-2013 110
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
20. Remainder term for Taylor series.
School of Mathematical Sciences Monash University
20.1 Motivation
It must be admitted that our approach to developing a power series expansion for func-
tions is somewhat cavalier. The weak spot in our analysis is that we have not given the
innite series due respect. We saw earlier that it is easy to obtain non-sensical results if
we make (rash) assumptions about the convergence of an innite series. These are deli-
cate areas and we must tread carefully. In this section we will use repeated integration
by parts to uncover each term in the power series. At each stage in the analysis we will
be working only with a nite series thus avoiding the delicate issues of convergence. As
an added bonus we will obtain, almost for free, an estimate of the error in using the
nite series as an approximation to the function. Excited? You should be.
20.2 Integration by parts and Taylor series
To get the ball rolling we will assume that we are given a function f(x) for which all
of its derivatives exist in some interval centred on x = 0, say R < x < R. That
is, f, f

, f

etc. are nite for every x in R < x < R. Now we start with the trivial
equation
f(x) = f(0) +
_
x
0
f

(u) du
This is nothing more than the familiar rule for denite integrals (more correctly known
as the Fundamental Theorem of Integral Calculus). The integral on the right will now
be manipulated by way of an integration by parts. To avoid confusion we will write the
usual rule for integration by parts in the non-standard form
_
p

(u)q(u) du = p(u)q(u) +
_
p(u)q

(u) du
For our case we will choose
p

(u) = 1 q(u) = f

(u)
Now we need to compute p(u), the anti-derivate of p

(u) = 1 and q

(u) the derivative


of q(u). In integration by parts we would not normally include an integration constant
when computing p(u) from p

(u) but on this occasion we will make a careful choice (the


reason will become obvious soon). So we choose
p(u) = (u x) q

(u) = f

(u)
and this leads to
f(x) = f(0) +
_
(u x)f

(u)
_
u=x
u=0
+
_
x
0
(u x)f

(u) du
= f(0) + xf

(0) +
_
x
0
(u x)f

(u) du
26-Jul-2013 112
School of Mathematical Sciences Monash University
We can now repeat the integration by parts, this time on the new and apparently more
complicated integral. This time we choose
p

(u) = (u x) q(u) = f

(u)
p(u) =
1
2
(u x)
2
q

(u) = f

(u)
which leads to
f(x) = f(0) + xf

(0) +
_

1
2
(u x)
2
f

(u)
_
u=x
u=0
+
_
x
0
1
2
(u x)
2
f

(u) du
= f(0) + xf

(0) +
x
2
2
f

(0) +
_
x
0
1
2
(u x)
2
f

(u) du
Just so that we all understand the procedure, we will do one more round of the integration
by parts. Here we choose
p

(u) =
1
2
(u x)
2
q(u) = f

(u)
p(u) =
1
3!
(u x)
3
q

(u) = f
iv
(u)
leading to
f(x) = f(0) + xf

(0) +
x
2
2
f

(0) +
_

1
3!
(u x)
3
f

(u)
_
u=x
u=0
+
_
x
0
1
3!
(u x)
3
f
iv
(u) du
= f(0) + xf

(0) +
x
2
2
f

(0) +
_
x
0
1
3!
(u x)
3
f
iv
(u) du
Its now rather obvious how the equations unfold with further rounds of integrations by
parts.
Maclaurin series with remainder term
Given a function f(x) innitely dierentiable on the interval R < x < R then
f(x) = P
n
(x) + E
n
(x) R < x < R
with
P
n
(x) = f(0) + xf

(0) +
x
2
2!
f

(0) + +
x
n
n!
f
(n)
(0)
E
n
(x) =
_
x
0
(1)
n
(u x)
n
n!
f
(n+1)
(u) du
Of course there will be many functions for which an expansion around x = 0 is not
appropriate (for example f(x) = 1/x. In such cases we would be forced to choose some
26-Jul-2013 113
School of Mathematical Sciences Monash University
other point (say x = 1 for f(x) = 1/x) on which to build the power series. It is a simple
matter to adapt the method just given for the Maclaurin series to this more general
case. The result is as follows
Taylor series with remainder term
Given a function f(x) innitely dierentiable on the interval R < x a < R then
f(x) = P
n
(x) + E
n
(x) R < x a < R
with
P
n
(x) = f(a) + (x a)f

(a) +
(x a)
2
2!
f

(a) + +
(x a)
n
n!
f
(n)
(a)
E
n
(x) =
_
x
a
(1)
n
(u x)
n
n!
f
(n+1)
(u) du
The polynomial P
n
(x) is commonly used as an approximation to f(x). The term E
n
(x)
is known as the remainder term. Later we shall see how we can use estimates of E
n
(x)
to not only provide a bound on the error incurred in using P
n
(x) as an approximation
to f(x) but also as another tool in deciding if the innite series converges or diverges.
You might wonder if we have made any progress. We certainly have. No longer must
we worry about the problems that come with working directly on an innite series. At
every stage in our analysis we have just three functions, f(x), a polynomial P
n
(x) and
a remainder term E
n
(x). Each of these functions can be manipulated using any of the
tools of algebra or calculus. This is a good place to be. From this solid foundation we
can carefully recover the innite series by looking at E
n
(x) in the limit as n .
As an example, consider the function f(x) = e
x
which we will expand about the point
x = 0. Following the steps given above we would arrive at
e
x
= P
n
(x) + E
n
(x)
with
P
n
(x) = 1 +
x
1!
+
x
2
2!
+
x
3
3!
+ +
x
n
n!
E
n
(x) =
_
x
0
(1)
n
(u x)
n
n!
e
u
du
We can now ask (and answer) two interesting questions
For a given n, how large might the error be in using P
n
(x) as an approximation to
e
x
?
For what values of x will the sequence P
n
(x), n = 1, 2, 3, converge?
26-Jul-2013 114
School of Mathematical Sciences Monash University
The answers can be found in a careful study of the remainder term E
n
(x). Let us start
with the rst question (how curious). Since e
x
= P
n
(x) + E
n
(x) we see that
|e
x
P
n
(x)| = |E
n
(x)| =
_
x
0
(u x)
n
n!
e
u
du
It is true that we could, after some eort, evaluate the right hand side to obtain the
exact size of the error |e
x
P
n
(x)| but in most cases this is not easy to do (e.g.. try using
f(x) = e
x
2
). So the usual approach here is to make some simple assumptions about the
size of the terms in the integral so that we can obtain an equation of the form
|e
x
P
n
(x)| =
_
x
0
(u x)
n
n!
e
u
du < M
n
(x)
for some M
n
(x). In this way M
n
(x) serves an estimate for the error in using P
n
(x) as
an approximation for e
x
. Note that the particular form for M
n
(x) will depend on the
assumptions we make in simplifying the messy integral, dierent assumptions will lead
to dierent functions M
n
(x). Obviously our interest is nding a good estimate M
n
(x).
Okay, enough of the game plan, lets get back to our example. Since we know that the
maximum for e
x
over the interval R < x < R is e
R
we have
_
x
0
(u x)
n
n!
e
u
du <
_
x
0
(u x)
n
n!
e
R
du
The right hand side is now a trivial integral and we obtain
_
x
0
(u x)
n
n!
e
u
du < e
R
x
n+1
(n + 1)!
= M
n
(x)
Thus we have
|e
x
P
n
(x)| < e
R
x
n+1
(n + 1)!
This gives us an answer to our rst question. That is, the error in using P
n
(x) as an
approximation to e
x
is no larger than e
R
x
n+1
/(n + 1)! for R < x < R. It is also
gives us the tools needed to answer the second question. Notice that for any xed x in
R < x < R we have
lim
n
e
R
x
n+1
(n + 1)!
= 0
In other words,
e
x
= lim
n
P
n
(x)
for R < x < R. And as there is no restriction on R we see that we can allow R
and still guarantee convergence. In short, we have just shown that the innite series
1 +
x
1!
+
x
2
2!
+
x
3
3!
+ +
x
n
n!
+
converges to e
x
for any x and the radius of convergence is .
26-Jul-2013 115
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
21. Introduction to ODEs
School of Mathematical Sciences Monash University
21.1 Motivation
The mathematical description of the real world is most commonly expressed in equations
that involve not just a function f(x) but also some of its derivatives. These equations
are known as ordinary dierential equations (commonly abbreviated as ODEs). Here
are some typical examples.
Newtonian gravity m
d
2
r(t)
dt
2
=
GMm
r
2
Population growth
dN(t)
dt
= N(t)
Hanging chain
d
2
y(x)
dx
2
=
2
y(x)
Electrical currents L
dI(t)
dt
+ RI(t) = E sin(t)
The challenge for us is to nd the functions that are solutions to these equations. The
problem is that there is no systematic way to solve an ODE; thus we are forced to look
at a range of strategies. This will be our game for the next few lectures. We will identify
broad classes of ODES and develop particular strategies for each class.
21.2 Denitions
Here are some terms commonly used in discussions on ODEs.
Order The order of an ODE is the order of the highest derivative in the
ODE.
Linear The ODE only contains terms linear in the function and its deriva-
tives.
Non-linear Any ODE that is not a linear ODE.
Linear homogeneous A linear ODE that allows y = 0 as a solution.
Dependent variable The solution of the ODE. Usually y.
Independent variable The variable that the solution of the ODE depends
on. Usually x or t.
Boundary conditions A set of conditions that selects a unique solution
of the ODE. Essential for numerical work.
Initial value problem An ODE with boundary conditions given at a sin-
gle point. Usually found in time dependent prob-
lems.
26-Jul-2013 117
School of Mathematical Sciences Monash University
Boundary value problem An ODE with boundary conditions specied at
more than one point. Common in engineering prob-
lems.
Here are some typical ODEs (some of which we will solve in later lectures).
Linear rst order homogeneous
cos(x)
dy
dx
+ sin(x)y(x) = 0
Linear rst order non-homogeneous
cos(x)
dy
dx
+ sin(x)y(x) = e
2x
Non-linear second order
d
2
y
dx
2
+
_
dy
dx
_
2
+ y(x) = 0
Initial value problem
dN
dt
= 2N(t) , N(0) = 123
Boundary value problem
d
2
y
dx
2
+ 2
_
dy
dx
_
2
y(x) = 0 , y(0) = 0 , y(1) =
21.3 Solution strategies
There are at least three dierent approaches to solving ODEs and initial/boundary value
problems.
Graphical This uses a graphical means, where the value of dy/dx are inter-
preted as a direction eld, to trace out a particular solution of the
ODE. Primarily used for initial value problems.
Numerical Here we use a computer to solve the ODE. This is a very powerful
approach as it allows us to tackle ODEs not amenable to any other
approach. Used primarily for initial and boundary value problems.
26-Jul-2013 118
School of Mathematical Sciences Monash University
Analytical A full frontal assault with all the mathematical machinery we can
muster. This approach is essential if you need to nd the full
general solution of the ODE.
In this unit we will conne our attention to the last strategy, leaving numerical and
graphical methods for another day (no point over indulging on these nice treats).
So lets get this show on the road with a simple example.
Example 21.1
Find all functions y(x) which obey
0 =
dy
dx
+ 2x
First we rewrite the ODE as
dy
dx
= 2x
then we integrate both sides with respect to x
_
dy
dx
dx = 2
_
x dx
But _
dy
dx
dx =
_
dy = y(x) C
for any function y(x) and C is an arbitrary constant. Thus we have found
y(x) = C x
2
is a solution of the ODE for any choice of constant C. All solutions of the ODE must
be of this form (for a suitable choice of C).
Example 21.2
Find all functions y(x) such that
0 =
dy
dx
+ 2xy
If we proceed as before we might arrive at
_
dy
dx
dx = 2
_
xy dx
The left hand side is easy to evaluate but the right hand side is problematic we can
not easily compute its anti-derivative (we dont yet know y(x)). So we need a dierent
approach. This time we shue the y onto the left hand side,
_
1
y
dy
dx
dx = 2
_
x dx
26-Jul-2013 119
School of Mathematical Sciences Monash University
But _
1
y
dy
dx
dx =
_
1
y
dy = C + log y
thus we nd
log y = C x
2
y(x) = Ae
x
2
We succeeded in this example because we were able to shue all x terms to one side of
the equation and all y terms to the other. This is an example of a separable equation.
We shall meet these equations again in later lectures.
In both of these example we found that one constant of integration popped up. This
means that we found not one solution but a whole family, each member having a dierent
value for C. This family of solutions is often called the general solution of the ODE.
The role of boundary conditions (if given) is to allow a single member of the family to
be chosen.
21.4 General and particular solutions
Each time we take an anti-derivative, one constant of integration pops up. For a rst
order ODE we will need one anti-derivative and thus one constant of integration. But for,
say, a third order equation, we will need to apply three anti-derivatives, each providing
one constant of integration. What is the point of this discussion? It is the key to spotting
when you have found all solutions of the ODE. This is what you need to know.
General solution of an ODE
If y(x) is a solution of an nth order ODE and if y(x) contains n independent
integration constants then y(x) is the general solution of the ODE. Every solution
of the ODE will be found in this family.
Particular solution of an ODE
If y(x) is a solution of an nth order ODE and if y(x) contains no free constants,
then y(x) is a particular solution of the ODE.
Such solutions usually arise after the boundary conditions have been applied to the
general solution.
The great logician Betrand Russell once claimed that he could prove anything
if given that 1+1=1. So one day, some smarty-pants asked him, Ok. Prove
that youre the Pope. He thought for a while and proclaimed, I am one.
The Pope is one. Therefore, the Pope and I are one.
26-Jul-2013 120
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
22. Separable rst order ODEs.
School of Mathematical Sciences Monash University
22.1 Separable equations
In an earlier example we solved
dy
dx
=
x
y
by rst rearranging the equation so that y appeared only on the left hand side while x
appeared only on the right hand side. Thus we found
_
y dy =
_
x dx
and upon completing the integral for both sides we found
y
2
(x) = C x
2
This approach is known as separation of variables. It can only be applied to those ODEs
that allow us to shue the x and y terms onto separate sides of the ODE.
Separation of variables
If an ODE can be written in the form
dy
dx
=
f(x)
g(y)
then the ODE is said to be separable and its solution may be found from
_
g(y) dy =
_
f(x) dx
Example 22.1
Show that the ODE
e
x
dy
dx
2y = 1
is separable. Hence solve the ODE.
26-Jul-2013 122
School of Mathematical Sciences Monash University
Example 22.2
Show that
sin(x)
dy
dx
+ y
2
= cos(x)
is not separable.
Example 22.3
The number of bacteria in a colony is believed to grow according to the ODE
dN
dt
= 2N
where N(t) is the number of bacteria at time t. Given that N = 20 initially, nd N at
later times.
Example 22.4 : Newtons law of cooling
This is a simple model of how the temperature of a warm body changes with time.
The rate of change of the bodys temperature is proportional to the dierence between
the ambient and body temperatures. Write down a dierential equation that represents
this model and then solve the ODE.
Example 22.5
Use the substitution u(x) = x
2
+ y(x) to reduce the non-separable ODE
du
dx
= 3x
u
x
to a separable ODE. Hence obtain the general solution for u(x).
22.2 First order linear ODEs
This is a class of ODEs of the form
dy
dx
+ P(x)y = Q(x)
where P(x) and Q(x) are known functions of x.
We will study two strategies to solve such ODEs.
26-Jul-2013 123
School of Mathematical Sciences Monash University
Example 22.6
Given
dy
dx
+
1
x
y = 0
nd y(x).
For this ODE we have P(x) = 1/x and Q(x) = 0.
Solving this ODE is easy its a separable ODE, thus we have
dy
y
=
1
x
dx
and after integrating both sides we nd
y(x) =
C
x
where C is a (modied) constant of integration.
Note that the above ODE has y(x) = 0 as a particular solution.
Whenever a linear ODE has y(x) = 0 as a solution we say that the ODE is homogeneous.
Example 22.7
Show that y(x) = x is a particular solution of
dy
dx
+
1
x
y = 2
We call it a particular solution because it does not contain an arbitrary constant of
integration.
This ODE looks very much like the previous example with the one small change that
Q(x) = 2 rather than Q(x) = 0. We can expect that the general solution will be similar
to the solution found in the previous example.
26-Jul-2013 124
School of Mathematical Sciences Monash University
Example 22.8
Show that
y(x) =
C
x
+ x
is the general solution of the ODE in the previous example.
Thus we have solved the ODE by a two step process, rst by solving the homogeneous
equation and second by nding any particular solution.
Strategy 1 for Linear 1st Order ODEs
Suppose that y
h
(x) is the general solution of the homogeneous equation
dy
h
dx
+ P(x)y
h
= 0
and suppose that y
p
(x) is any particular solution of
dy
dx
+ P(x)y = Q(x)
Then the general solution of the previous ODE is
y(x) = y
h
(x) + y
p
(x)
Note, in some books y
h
(x) is written as y
c
(x) and is known as the complementary solu-
tion.
Though this above procedure sounds easy we still have two problems,
How do we compute the general solution of the homogeneous ODE?
How do we obtain a particular solution?
22.2.1 Solving the homogeneous ODE
Setting Q(x) = 0 leads to
dy
dx
+ P(x)y = 0
This is separable, thus we have
26-Jul-2013 125
School of Mathematical Sciences Monash University
dy
y
= P(x)dx
which we can integrate, with the result
y(x) = Ce

P(x) dx
Remember that this y(x) will be used as y
h
(x), the homogeneous solution of the non-
homogeneous ODE.
Example 22.9
Verify the above solution for y(x)
22.2.2 Finding a particular solution
This usually involves some inspired guess work. The general idea is to look at Q(x)
and then guess a class of functions for y
p
(x) that might be a solution of the ODE. If
you include few free parameters you may be able to nd a particular solution any
particular solution will do.
Pi goes on and on and on ...
And e is just as cursed
I wonder: Which is larger
When their digits are reversed?
26-Jul-2013 126
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
23. The integrating factor.
School of Mathematical Sciences Monash University
23.1 The Integrating Factor
Example 23.1 : Easy
Use an inspired guess to nd a particular solution of
dy
dx
+ 3y = sin(x)
Example 23.2 : Harder
Use an inspired guess to nd a particular solution of
dy
dx
+ (1 + 3x)y = 3e
x
The main advantage of this method of inspired guessing (better known as the method of
undetermined coecients) is that it is easy to apply. The main disadvantage is that it is
not systematic it involves an element of guess work in nding the particular solution.
We need another, better and systematic strategy.
We begin by noticing that for any function I(x),
1
I
d(Iy)
dx
=
dy
dx
+ y
1
I
dI
dx
The right hand side looks similar to the left hand side of our generic rst order linear
ODE. We can make it exactly the same by choosing I(x) such that
P(x) =
1
I
dI
dx
This is a separable ODE for I(x), with the particular solution
I(x) = e

P(x) dx
So why our we doing this? Because once we know I(x) our original ODE may be re-
written as
1
I
d(Iy)
dx
= Q(x)
We can now integrate this,
26-Jul-2013 128
School of Mathematical Sciences Monash University
d(Iy)
dx
= I(x)Q(x)

_
d(Iy)
dx
dx =
_
I(x)Q(x) dx
I(x)y(x) =
_
I(x)Q(x) dx
y(x) =
1
I(x)
_
I(x)Q(x) dx
The great advantage with this method is that it works every time! No guessing!
The function I(x) is known as the integrating factor.
Strategy 2 for Linear 1st Order ODEs
The general solution of
dy
dx
+ P(x)y = Q(x)
is
y(x) =
1
I(x)
_
I(x)Q(x) dx
where the integrating factor I(x) is given by
I(x) = e

P(x) dx
Example 23.3
Find the general solution of
dy
dx
+
1
x
y = 2
Here we have P(x) = 1/x and Q(x) = 2.
26-Jul-2013 129
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
24. Homogeneous Second order ODEs.
School of Mathematical Sciences Monash University
24.1 Second order linear ODEs
The most general second order linear ODE is
P(x)
d
2
y
dx
2
+ Q(x)
dy
dx
+ R(x)y = S(x)
Such a beast is not easy to solve. So we are going to make life easy for ourselves by
assuming P(x), Q(x), R(x) and S(x) are constants. Thus we will be studying the
reduced class of linear second order ODEs of the form
a
d
2
y
dx
2
+ b
dy
dx
+ cy = S(x)
where a, b, and c are constants.
No prizes for guessing that these are called constant coecient equations.
We will consider two separate cases, the homogeneous equation where S(x) = 0 and the
non-homogeneous equation where S(x) = 0.
24.2 Homogeneous equations
Here we are trying to nd all functions y(x) that are solutions of
a
d
2
y
dx
2
+ b
dy
dx
+ cy = 0
Lets take a guess, lets try
y(x) = e
x
We introduce the parameter as something to juggle in the hope that y(x) can be made
to be a solution of the ODE. First we need the derivatives,
y(x) = e
x

dy
dx
= e
x

d
2
y
dx
2
=
2
e
x
Then we substitute this into the ODE
0 = a
2
e
x
+ be
x
+ ce
x
0 = (a
2
+ b + c)e
x
but e
x
= 0
0 = a
2
+ b + c
26-Jul-2013 131
School of Mathematical Sciences Monash University
So we have a quadratic equation for , its two solutions are

1
=
b +

b
2
4ac
2a
and
2
=
b

b
2
4ac
2a
Lets assume for the moment that
1
=
2
and that they are both real numbers.
What does this all mean? Simply that we have found two distinct solutions of the ODE,
y
1
(x) = e
1x
and y
2
(x) = e
2x
Now we can use two of the properties of the ODE, one, that it is linear and two, that it
is homogeneous, to declare that
y(x) = Ay
1
(x) + By
2
(x)
is also a solution of the ODE for any choice of constants A and B.
Example 24.1
Prove the previous claim, that y(x) is a solution of the linear homogeneous ODE.
And now comes the great moment of enlightenment the y(x) just given contains two
arbitrary constants and as the general solution of a second order ODE must contain two
arbitrary constants we now realise that y(x) above is the general solution.
Example 24.2 : Real and distinct roots
Find the general solution of
d
2
y
dx
2
+
dy
dx
6y = 0
First we solve the quadratic

2
+ 6 = 0
for . This gives
1
= 2 and
2
= 3 and thus
y(x) = Ae
2x
+ Be
3x
is the general solution.
The quadratic equation
26-Jul-2013 132
School of Mathematical Sciences Monash University
a
2
+ b + c = 0
arising from the guess y(x) = e
x
is known as the characteristic equation for the ODE.
We have already studied one case where the two roots are real and distinct. Now we
shall look at some examples where the roots are neither real nor distinct.
Example 24.3 : Complex roots
Find the general solution of
d
2
y
dx
2
2
dy
dx
+ 5y = 0
First we solve the quadratic

2
2 + 5 = 0
for . This gives
1
= 1 2i and
2
= 1 + 2i. These are distinct but they are complex.
Thats not a mistake just a venture into slightly unfamiliar territory. The full solution
is still given by
y(x) = Ae
1x
+ Be
2x
= Ae
(12i)x
+ Be
(1+2i)x
This is a perfectly correct mathematical expression and it is the solution of the ODE.
However, in cases where the solution of the ODE is to be used in a real-world problem,
we would expect y(x) to be a real-valued function of the real variable x. In such cases
we must therefore have both A and B as complex numbers. This is getting a bit messy
so its common practice to re-write the general solution as follows.
First recall that e
i
= cos + i sin and thus
e
(12i)x
= e
x
(cos(2x) i sin(2x))
e
(1+2i)x
= e
x
(cos(2x) + i sin(2x))
and thus our general solution is also
y(x) = e
x
((A + B) cos(2x) + (iA + iB) sin(2x))
Now A + B and iA + iB are constants so lets just replace them with a new A and a
new B, that is we write
y(x) = e
x
(Acos(2x) + Bsin(2x))
This the general solution of the ODE written in a form suitable for use with real numbers.
26-Jul-2013 133
School of Mathematical Sciences Monash University
Example 24.4 : Equal roots
Find the general solution of
d
2
y
dx
2
+ 2
dy
dx
+ y = 0
This time we nd just one root for ,

1
=
2
= 1
If we tried to declare that
y(x) = Ae
1x
+ Be
2x
= Ae
x
+ Be
x
was the general solution we would be fooling ourselves. Why? Because in this case the
two integration constants combine into one
y(x) = Ae
x
+ Be
x
= (A + B)e
x
= Ce
x
where C = A + B. We need two independent constants in order to have a general
solution.
The trick in this case is this. Put
y(x) = (A + Bx)e
x
This does have two independent constants and you can show that this is a solution of
the ODE for any choice of A and B. Thus it must also be the general solution.
The upshot of all of this is that when solving the general linear second order homogeneous
ODE we have three cases to consider, real and distinct roots, complex root and equal
roots. The recipe to apply in each case is listed in the following table.
26-Jul-2013 134
School of Mathematical Sciences Monash University
Constant coecient 2nd order homogeneous ODEs
For the ODE
a
d
2
y
dx
2
+ b
dy
dx
+ cy = 0
rst solve the quadratic
a
2
+ b + c = 0
for . Let the two roots be
1
and
2
. Then for the general solution of the previous
ODE there are three cases.
Case 1 :
1
=
2
and real y(x) = Ae
1x
+ Be
2x
Case 2 : = i y(x) = e
x
(Acos(x) + Bsin(x))
Case 3 :
1
=
2
y(x) = (A + Bx)e
x
26-Jul-2013 135
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
25. Non-Homogeneous Second order ODEs.
School of Mathematical Sciences Monash University
25.1 Non-homogeneous equations
This is what the typical non-homogeneous linear constant coecient second order ordi-
nary dierential equation (phew!) looks like
a
d
2
y
dx
2
+ b
dy
dx
+ cy = S(x)
where a, b, c are constants and S(x) = 0 is some given function. This diers from the
homogeneous case only in that here we have S(x) = 0.
Our solution strategy is very similar to that which we used on the general linear rst
order equation. There we wrote the general solution as
y(x) = y
h
(x) + y
p
(x)
where y
h
is the general solution of the homogeneous equation and y
p
(x) is any particular
solution of the ODE.
We will use this same strategy for solving our non-homogeneous 2nd order ODE.
Example 25.1
Find the general solution of
d
2
y
dx
2
+
dy
dx
6y = 1 + 2x
This proceeds in three steps, rst, solve the homogeneous problem, second, nd a par-
ticular solution and third, add the two solutions together.
Step 1 : The homogeneous solution
Here we must nd the general solution of
d
2
y
h
dx
2
+
dy
h
dx
6y
h
= 0
for y
h
. In the previous lecture we found
y
h
(x) = Ae
2x
+ Be
3x
Step 2 : The particular solution
Here we have to nd any solution of the original ODE. Since the right hand side is a
polynomial we try a guess of the form
y
p
(x) = a + bx
26-Jul-2013 137
School of Mathematical Sciences Monash University
where a and b are numbers (which we have to compute).
Substitute this into the left hand side of the ODE and we nd
d
2
(a + bx)
dx
2
+
d(a + bx)
dx
6(a + bx) = 1 + 2x
b 6a 6bx = 1 + 2x
This must be true for all x and so we must have
b 6a = 1 and 6b = 2
from which we get b = 1/3 and a = 2/9 and thus
y
p
(x) =
2
9

1
3
x
Note nding a particular solution be this guessing method is often called the method of
undetermined coecients.
Step 3 : The general solution
This is the easy bit
y(x) = y
h
(x) + y
p
(x) = Ae
2x
+ Be
3x

2
9

1
3
x
Our job is done!
25.2 Undetermined coecients
How do we choose a workable guess for the particular solution? Simply by inspecting
the terms in S(x), the right hand side of the ODE.
Here are some examples,
Guessing a particular solution
S(x) = (a + bx + cx
2
+ + dx
n
)e
kx
try y
p
(x) = (e + fx + gx
2
+ + hx
n
)e
kx
S(x) = (a sin(bx) + c cos(bx))e
kx
try y
p
(x) = (c cos(bx) + f sin(bx))e
kx
26-Jul-2013 138
School of Mathematical Sciences Monash University
Example 25.2
What guesses would you make for each of the following?
S(x) = 2 + 7x
2
S(x) = (sin(2x))e
3x
S(x) = 2x + 3x
3
+ sin(4x) 2xe
3x
25.3 Exceptions
Without exception there are always exceptions!
If S(x) contains terms that are solutions of the corresponding homogeneous equation
then in forming the guess for the particular solution you should multiply that term by x
(and by x
2
if the term corresponded to a repeated root of the characteristic equation).
Example 25.3
Find the general solution for
d
2
y
dx
2
+
dy
dx
6y = e
2x
The homogeneous solution is
y
h
(x) = Ae
2x
+ Be
3x
and thus we see that our right hand side contains a piece of the homogeneous solution.
The guess for the particular solution would then be
y
p
(x) = (a + bx)e
2x
Now solve for a and b.
26-Jul-2013 139
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
26. Coupled systems of ODEs
School of Mathematical Sciences Monash University
26.1 Motivation
Are we enjoying this game of solving dierential equations? Well here is change of
scenery that should keep your little grey cells ticking over.
Previously we have been studying simple equations for one unknown (usually y(x)). Now
we shall up the ante by considering equations like the following
du
dx
= 6u + 16v (A)
dv
dx
= u 4v (B)
This is a coupled system of ODEs and the challenge now is to solve this system for the
two functions u(x) and v(x). One popular example of this kind of coupled system occurs
in the study of competing populations, commonly called a predator-prey model. In this
case u(x) and v(x) would record the number of predator and prey as functions of time,
uncommonly and unfashionably recorded as x.
How might we solve this system? As they stand we are in a bit of a bind. The functions
u(x) and v(x) appear in both dierential equations. If we are to make progress it seems
reasonable then to apply some black-magic (soon to be fully revealed) that decouples
these equations. That is, we seek to obtain related equations, each containing only one
unknown function, and each in a form where we can employ familiar techniques to obtain
the solutions. As we shall soon see, there are at least two roads we can follow, one uses
dierentiation to decouple the equations, the other uses eigenvector methods. Both have
their merits.
Let the games begin.
26.2 First method: dierentiation
Suppose that we chose (for fun?) to dierentiate the two equations, the result would be
d
2
u
dx
2
= 6
du
dx
+ 16
dv
dx
d
2
v
dx
2
=
du
dx
4
dv
dx
The rst derivatives on the right hand side can be replaced using the original equations
(A) and (B) and this leads to
d
2
u
dx
2
= 20u + 32v (C)
d
2
v
dx
2
= 2u (D)
26-Jul-2013 141
School of Mathematical Sciences Monash University
It might seem that we are going backwards as we still have a coupled system and, quelle
horror, we now have to contend with second derivatives. But there is light at the end
of the tunnel. Look carefully at equations (A) and (C) and you will notice that the
combination (C) 2(A) will eliminate v(x) from the right hand side. Likewise, the
combination (D) 2(B) will eliminate u(x) from the right hand side. Here is what we
get,
(C) 2(A)
d
2
u
dx
2
2
du
dx
8u = 0 (E)
(D) 2(B)
d
2
v
dx
2
2
dv
dx
8v = 0 (F)
Now we are happy, we have two equations with one unknown in each equation. These
equations are easy to solve, namely
u(x) = Ae
4x
+ Be
2x
(G)
v(x) = Ce
4x
+ De
2x
(H)
We might be tempted to declare that this pair of equations dene the full general solution
of the original ODEs. But we have a minor problem. In this so-called general solution
we have four integration constants, A, B, C and D (not to be confused with the the
equation labels). However, we started with just two rst order ODEs and so we expect
just two integration constants, not four. Somehow we have to reduce the number of
constants from four to two. Here is one way to do just that, substitute the solutions (G)
and (H) back into the original ODE (A). This leads to
4Ae
4x
2Be
2x
= 6
_
Ae
4x
+ Be
2x
_
+ 16
_
Ce
4x
+ De
2x
_
= (6A + 16C) e
4x
+ (6B + 16D) e
2x
This must be true for all x, thus we can equate the coecients of corresponding expo-
nential terms leading to
4A = 6A + 16C A = 8C
2B = 6B + 16D B = 2D
Thus we have
u(x) = 8Ce
4x
2De
2x
v(x) = Ce
4x
+ Dd
2x
Example 26.1
Use the solution for u(x) given by (G) to compute v(x) directly from equation (A).
26-Jul-2013 142
School of Mathematical Sciences Monash University
Example 26.2
Rather than substituting (G) and (H) into (A) try substituting into (B). Do you get
the same solutions?
26.3 Second method: eigenvectors and eigenvalues
The title of this section gives a clue as to where we are headed down the road of
eigenvectors and eigenvalues. You might think this to be a bit strange and you could
reasonably ask: What have eigenvectors got to do with dierential equations? The
surprise is that we have already encountered some of the key elements of the eigenvector
analysis. How so? Well, here are the main equations dressed up in matrix notation, rst
the ODEs,
_

_
du
dx
dv
dx
_

_
=
_
6 16
1 4
_ _
u
v
_
and second, the full solution
_
u(x)
v(x)
_
= Ce
4x
_
8
1
_
+ De
2x
_
2
1
_
Notice the two column vectors on the right hand side of the last equation? They happen
to be eigenvectors of the 2 2 matrix in the above (matrix) equation for the ODEs.
But wait, theres more. The terms e
4x
and e
2x
were found (in the previous section) by
solving the characteristic equation
0 =
2
2 8
with roots = 4 and = 2. That equation is also the equation for the eigenvalues of
the above 2 2 matrix. It is all coming together (agreed?).
Is all of this just pure coincidence? Not at all, and to prove the point we will now solve
a new set of coupled ODEs purely by matrix methods. Here is our new set of equations
_

_
du
dx
dv
dx
_

_
=
_
5 2
4 1
_ _
u
v
_
Now we compute the eigenvectors and eigenvalues of the 2 2 matrix
_
5 2
4 1
_
26-Jul-2013 143
School of Mathematical Sciences Monash University
The eigenvalues are = 3 and = 1 and the corresponding eigenvectors are
For = 3 v
1
=
_
1
1
_
For = 1 v
2
=
_
1
2
_
Example 26.3
Verify that the eigenvectors and eigenvalues are as stated.
Now here comes an important step (the rst of two important steps). We can write any
column vector as a linear combination of the two eigenvectors (in this instance). That
is, for any pair of numbers u and v (these will later be our functions u(x) and v(x)) we
can always write
_
u
v
_
= P
_
1
1
_
+ Q
_
1
2
_
for some numbers P and Q. How can we be so sure? Well, rst notice that this equation
is identical to the matrix equation
_
u
v
_
=
_
1 1
1 2
_ _
P
Q
_
This is a 22 matrix equation. Given any P, Q we can compute u, v. Likewise, given any
u, v we can compute P, Q (because the determinant of the coecient matrix is non-zero
and so the equations have a unique solution).
Now back to to our ODEs. We know that u(x) and v(x) are functions of x, so we will
now allow P and Q to be functions of x (remember, P and Q are totally arbitrary so
we are free to let them be functions of x). Thus we write
_
u(x)
v(x)
_
= P(x)
_
1
1
_
+ Q(x)
_
1
2
_
Next, substitute these into the (matrix) ODEs,
dP
dx
_
1
1
_
+
dQ
dx
_
1
2
_
=
_
5 2
4 1
_
_
P(x)
_
1
1
_
+ Q(x)
_
1
2
__
then carry the 2 2 matrix through the ( ) brackets on the right hand side to obtain
dP
dx
_
1
1
_
+
dQ
dx
_
1
2
_
= 3 P(x)
_
1
1
_
+ 1 Q(x)
_
1
2
_
Note how we have made use of the fact that the vectors inside the brackets were eigen-
vectors of the 2 2 matrix. Comparing coecients of the two column vectors on each
26-Jul-2013 144
School of Mathematical Sciences Monash University
side of this equation (this is the second important step) gives us the following pair of
ODEs
dP
dx
= 3P and
dQ
dx
= Q
This is a good. We now have two simple equations that are easily solved,
P(x) = Ae
3x
and Q(x) = Be
x
where A and B are integration constants. And so nally we have our full solution of the
coupled ODEs
_
u(x)
v(x)
_
= Ae
3x
_
1
1
_
+ Be
x
_
1
2
_
Example 26.4
Look carefully at the last step leading to the ODEs for P(x) and Q(x). Why were we
allowed to equate the coecients of the two column vectors on each side of the equation?
You might nd it instructive to shue all the terms onto one side of the equals sign and
then write those equations as a 2 2 matrix equation.
Example 26.5
Solve the above ODEs for u(x) and v(x) using the method given in the previous section.
Verify that you get the same solution as given above (as you would expect).
It has been a long road. You might wonder why we would take such a long journey when
a more direct method (as in the previous section using dierentiation and elimination)
is available. The answer is that it introduces you to a very powerful technique, the use
of eigenvectors as a basis for a vector space, that can greatly simplify the algebraic com-
plexity of many problems. This technique is used in many other areas of mathematics,
most commonly in linear algebra, but also, as in this case, in solving dierential equa-
tions. It is also the corner stone of quantum mechanics, the wave function of a system
is a linear combination of eigenfunctions (a variation on the theme of eigenvectors).
26-Jul-2013 145
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
27. Applications of Dierential Equations
School of Mathematical Sciences Monash University
27.1 Applications of ODEs
In the past few lectures we studied, in detail, various techniques for solving a wide
variety of dierential equations. What we did not do is ask why we would want to solve
those equations in the rst place. A simple (but rather weak) answer is that it is a nice
intellectual challenge. A far better answer is that these ODEs arise naturally in the
study of a vast array of physical problems, such as population dynamics, the spread of
infectious diseases, the cooling of warm bodies, the swinging motion of a pendulum and
the motion of planets. In this lecture we shall look at some of these applications.
In each of the following examples we will not spend time computing the solution of the
ODE this is left as an exercise for the (lucky) student!
27.2 Newtons law of cooling
Newtons law of cooling states that the rate of change of the temperature of a body is
directly proportional to the temperature dierence between the body and its surrounding
environment. Let the temperature of the body be T and let T
a
be that of the surrounding
environment (the ambient temperature). Then Newtons law of cooling is expressed in
mathematical terms as
dT
dt
= k(T T
a
)
where k is some constant.
This is a simple non-homogeneous rst order linear dierential equation. Its general
solution is
T(t) = T
a
+ Ae
kt
To apply this equation to a specic example we would need information that allows us
to assign numerical values to the three parameters, T
a
, k, and A.
Example 27.1 : A murder scene
We can use Newtons law of cooling to estimate the time of death at a murder scene.
Suppose the temperature of the body has been measured at 30 deg C. The normal body
temperature is 37 deg C. So the question is How long does it take for the body to cool
from 37 deg C to 30 deg C? To answer this we need values for T
a
, k, and A. Suppose
the room temperature was 20 deg C and thus T
a
= 20. For k we need to draw upon
previous experiments (how?). These show that a body left to cool in a 20 deg C room
will drop from 37 deg C to 35 deg C in 2 hours. Substitute this into the above equation
and we have
26-Jul-2013 147
School of Mathematical Sciences Monash University
T(0) = 37 = 20 + Ae
0
T(2) = 35 = 20 + Ae
2k
Two equations in two unknowns, A and k. These are easy to solve, leading to
A = 17 and k =
1
2
log
e
_
17
15
_
0.06258
Thus
T(t) = 20 + 17e
0.06258t
Now for the time of the murder. Put T(t) = 30 and solve for t,
30 = 20 + 17e
0.06258t
t =
1
0.06258
log
e
_
10
17
_
8.5
That is, the murder occurred about 8.5 hours earlier.
27.3 Pollution in swimming pools
Swimming pools should contain just two things people and pure water. Yet all too
often the water is not pure. One way of cleaning the pool would be to pump in fresh
water (at one point in the pool) while extracting the polluted water (at some other
point in the pool). Suppose we assume that the pools water remains thoroughly mixed
(despite one entry and exit point) and that the volume of water remains constant. Can
we predict how the level of pollution changes with time?
Suppose at time t there is y(t) kgs of pollutant in the pool and that the volume of the
pool is V litres. Suppose also that pure water is owing in at the rate litres/min and,
since the volume remains constant, the outow rate is also litres/min.
Now we will set up a dierential equation that describes how y(t) changes with time.
Consider a small time interval, from t to t + t, where t is a small number. In that
interval t litres of polluted water was extracted. How much pollutant did this carry?
As the water is uniformly mixed we conclude that the density of the pollutant in the
extracted water is the same as that in the pool. The density in the pool is y/V kg/L
and thus the amount of pollutant carried away was (y/V )(t). In the same small time
interval no new pollutants were added to the pool. Thus any change in y(t) occurs solely
from the ow of pollutants out of the pool. We thus have
y(t + t) y(t) =
y
V
t
26-Jul-2013 148
School of Mathematical Sciences Monash University
This can be reduced to a dierential equation by dividing through by t and then taking
the limit as t 0. The result is
dy
dt
=

V
y
The general solution is
y(t) = y(0)e
t/V
Example 27.2
Suppose the water pumps could empty the pool in one day. How long would it take to
halve the level of pollution?
27.4 Newtonian mechanics
The original application of ODEs was made by Newton (at the age of 22 in 1660) in the
study of how things move. He formulated a set of laws, Newtons laws of motion, one
of which states that the nett force acting on a body equals the mass of the body times
the bodies acceleration.
Let F

be the force and let r

(t) be the position vector of the body. Then the bodys


velocity and acceleration are dened by
v

(t) =
dr

dt
a

(t) =
dv

dt
=
d
2
r

dt
2
Then Newtons (second) law of motion may be written as
m
d
2
r

dt
2
= F

If we know the force acting on the object then we can treat this as a second order
ODE for the particles position r

(t). The usual method of solving this ODE is to write


r

(t) = x(t) i

+ y(t)j

+ z(t)k

and to re-write the above ODE as three separate ODEs,


one each for x(t), y(t) and z(t).
26-Jul-2013 149
School of Mathematical Sciences Monash University
m
d
2
x
dt
2
= F
x
m
d
2
y
dt
2
= F
y
m
d
2
z
dt
2
= F
z
where F
x
, F
y
, F
z
are the components of the force in the directions of the (x, y, z) axes,
F

= F
x
i

+ F
y
j

+ F
z
k

.
Example 27.3 : Planetary motion
Newton also put forward a theory of gravitation that there exists a universal force of
gravity, applicable to every lump of matter in the universe, that states that for any pair
of objects the force felt by each object is given by
F =
Gm
1
m
2
r
2
where m
1
and m
2
are the (gravitational) masses of the respective bodies, r is the distance
between the two bodies and G is a constant (known as the Newtonian gravitational
constant and by experiment is found to be 6.67310
11
Nm
2
/kg
2
). The force is directed
along the line connecting the two objects.
Consider the motion of the Earth around the Sun. Each body will feel a force of gravity
acting to pull the two together. Each body will move due to the action of the force
imposed upon it be the gravitational pull of its partner. However as the Sun is far more
massive than the Earth, the Sun will, to a very good approximation, remain stationary
while the Earth goes about its business.
We can thus make the reasonable assumptions that
The Sun does not move.
The Sun is located at the origin of our coordinate system, x = y = z = 0
The Earth orbits the Sun in the z = 0 plane.
Let r

(t) = x(t) i

+ y(t)j

be the position vector of the Earth. The force acting on the


Earth due to the gravitational pull of the Sun is then given by
F

=
GMm
r
2
r

26-Jul-2013 150
School of Mathematical Sciences Monash University
where r

is a unit vector parallel to r

, M is the mass of the Sun and m is the mass of


the Earth. The minus sign shows that the force if pulling the Earth toward the Sun.
The unit vector is easy to compute, r

= (xi

+ yj

)/r. Thus we have, nally,


m
d
2
x
dt
2
=
GMm
r
2
x
r
m
d
2
y
dt
2
=
GMm
r
2
y
r
This is a non-linear coupled system of ODEs these are not easy to solve, so we resort
to (more) simple approximations (in other Maths subjects!).
Example 27.4 : Simple Harmonic Motion
Many physical systems display an oscillatory behaviour, such as a swinging pendulum
or a hanging weight attached to a spring. It seems reasonable then to expect the sine
an cosine functions to appear in the description of these systems. So what type of
dierential equation might we expect to see for such oscillatory systems? Simply those
ODEs that have the sine and cosine functions as typical solutions. We saw in previous
lectures that the ODE
d
2
y
dt
2
= k
2
y
has
y(t) = Acos(kt) + Bsin(kt)
as its general solution. This the classic example of what is called simple harmonic
motion. Both the swinging pendulum and the weighted spring are described (actually
approximated) by the above simple harmonic equation.
26-Jul-2013 151
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
28. Functions of Several Variables
School of Mathematical Sciences Monash University
28.1 Introduction
We are all familiar with simple functions such as y = sin(x). And we all know the
answers (dont we?) to questions such as
1. What is its domain and the range ?
2. What does it look like as a plot in the xy-plane?
3. What is its derivative?
In this series of lectures we are going to up the ante by exploring similar questions for
functions similar to z = cos(xy). This is just one example of what we call functions
of several variables. Though we will focus on functions that involve three variables
(usually x, y and z) the lessons learnt here will be applicable to functions of any number
of variables.
28.2 Denition
A function f of two variables (x, y) is a single valued mapping of a subset of R
2
into a
subset of R.
What does this mean? Simply that for any allowed value of x and y we can compute a
single value for f(x, y). In a sense f is a process for converting pairs of numbers (x and
y) into a single number f.
The notation R
2
means all possible choices of x and y such as all points in the xy-plane.
The symbol R denotes all real numbers (for example all points on the real line). The
use of the word subset in the above denition is simply to remind us that functions have
an allowed domain (i.e. a subset of R
2
) and a corresponding range (i.e. a subset of R).
Notice that we are restricting ourselves to real variables, that is the functions value and
its arguments (x, y) are all real numbers. This game gets very exciting and somewhat
tricky when we enter the world of complex numbers. Such adventures await you in later
year mathematics (not surprisingly this area is known as Complex Analysis).
28.3 Notation
Here is a simple function of two variables
f(x, y) = sin(x + y)
We can choose the domain to be R
2
and then the range will be the closed set [1, +1].
Another common way of writing all of this is
f : (x, y) R
2
sin(x + y) [1, 1]
26-Jul-2013 153
School of Mathematical Sciences Monash University
This notation identies the function as f, the domain as R
2
, the range as [1, 1] and
most importantly the rule that (x, y) is mapped to sin(x + y). For this subject we will
stick with the former notation.
You should also note that there is nothing sacred about the symbols x, y and f. We
are free to choose what ever symbols takes our fancy, for example we could concoct the
function
w(u, v) = log(u v)
Example 28.1
What would be a sensible choice of domain for the previous function?
28.4 Surfaces
A very common application of a function of two variables is to describe a surface in
3-dimensional space. How so? you might ask. The idea is that we take the value of the
function to describe the height of the surface above the xy-plane. If we use standard
Cartesian coordinates then such a surface could be described by the equation
z = f(x, y)
This surface has a height z units above each point (x, y) in the xy-plane.
The equation z = f(x, y) describes the surface explicitly as a height function over a
plane and thus we say that the surface is given in explicit form.
A surface such as z = f(x, y) is also often called the graph of the function f.
Here are some simple examples. A very good exercise is to try to convince yourself that
the following images are correct (i.e. that they do represent the given equation).
Note that in each of the following r is dened as r = +
_
(x
2
+ y
2
).
26-Jul-2013 154
School of Mathematical Sciences Monash University
z = x
2
+ y
2
1 = x
2
+ y
2
z
2
z = cos (3r) exp (2r
2
)
26-Jul-2013 155
School of Mathematical Sciences Monash University
z =
_
1 + y
2
x
2
z = xy exp (x
2
y
2
)
1 = x + y + z
26-Jul-2013 156
School of Mathematical Sciences Monash University
28.5 Alternative forms
We might ask are there any other ways in which we can describe a surface? We should
be clear that (in this subject) when we say surface we are talking about a 2-dimensional
surface in our familiar 3-dimensional space. With that in mind, consider the equation
0 = g(x, y, z)
What do we make of this equation? Well, after some algebra we might be able to
re-arrange the above equation into the familiar form
z = f(x, y)
for some function f. In this form we see that we have a surface, and thus the previous
equation 0 = g(x, y, z) also describes a surface. When the surface is described by an
equation of the form 0 = g(x, y, z) we say that the surface is given in implicit form.
Consider all of the points in R
3
(i.e all possible (x, y, z) points). If we now introduce the
equation 0 = g(x, y, z) we are forced to consider only those (x, y, z) values that satisfy
this constraint. We could do so by, for example, arbitrarily choosing (x, y) and using
the equation (in the form z = f(x, y) to compute z. Or we could choose say (y, z) and
use the equation 0 = g(x, y, z) to compute x. Which ever road we travel it is clear that
we are free to choose just two of the (x, y, z) with the third constrained by the equation.
Now consider some simple surface and lets suppose we are able to drape a sheet of graph
paper over the surface. We can use this graph paper to select individual points on the
surface (well as far as the graph paper covers the surface). Suppose we label the axes
of the graph paper by the symbols u and v. Then each point on surface is described by
a unique pair of values (u, v). This makes sense we are dealing with a 2-dimensional
surface and so we expect we would need 2 numbers ((u, v)) to describe each point on the
surface. The parameters (u, v) are often referred to as (local) coordinates on the surface.
How does this picture t in with our previous description of a surface, as an equation of
the form 0 = g(x, y, z)? Pick any point on the surface. This point will have both (x, y, z)
and (u, v) coordinates. That means that we can describe the point in terms of either
(u, v) or (x, y, z). As we move around the surface all of these coordinates will vary. So
given (u, v) we should be able to compute the corresponding (x, y, z) values. That is we
should be able to nd functions P(u, v), Q(u, v) and R(u, v) such that
x = P(u, v) y = Q(u, v) z = R(u, v)
The above equations describe the surface in parametric form.
Example 28.2
Identify (i.e. describe) the surface given by the equations
x = 2u + 3v + 1 y = u 4v + 2 z = u + 2v 1
Hint : Try to combine the three equations into one equation involving x, y and z but
not u and v.
26-Jul-2013 157
School of Mathematical Sciences Monash University
Example 28.3
Describe the surface dened by the equations
x = 3 cos() sin() y = 4 sin() sin() z = 5 cos()
for 0 < < 2 and 0 < <
Example 28.4
How would your answer to the previous example change if the domain for was 0 < <
/2?
Equations for surfaces
A 2-dimensional surface in a 3-dimensional space may be described by any of the
following forms.
Explicit z = f(x, y)
Implicit 0 = g(x, y, z)
Parametric x = P(u, v), y = Q(u, v), z = R(u, v)
26-Jul-2013 158
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
29. Partial derivatives
School of Mathematical Sciences Monash University
29.1 First derivatives
We all know and love the familiar denition of a derivative of a function of one variable,
df
dx
= lim
x0
f(x + x) f(x)
x
The natural question to ask is : Is there similar rule for functions of more than one
variable? The answer is yes (surprised?) and we will develop the necessary formulas by
a simple generalisation of the above denition.
Okay, lets suppose we have a simple function, say f(x, y). Suppose for the moment that
we pick a particular value of y, say y = 3. Then only x is allowed to vary and in eect
we now have a function of just one variable. Thus we can apply the above denition for
a derivative which we write as
f
x
= lim
x0
f(x + x, y) f(x, y)
x
Notice the use of the symbol rather than d. This is to remind us that in computing
this derivative all other variables are held constant (which in this instance is just y).
Of course we could play the same again but with x held constant, this leads to derivative
in y,
f
y
= lim
y0
f(x, y + y) f(x, y)
y
Each of these derivatives, f/x and f/y are known as partial derivatives of f while
the derivative of a function of one variable is often called an ordinary derivative.
You might think that we would now need to invent new rules for the (partial) derivatives
of products, quotients and so on. But our denition of partial derivatives is built upon
the denition of an ordinary derivative of a function of one variable. Thus all the
familiar rules carry over without modication. For example, the product rule for partial
derivatives is
(fg)
x
= g
f
x
+ f
g
x
(fg)
y
= g
f
y
+ f
g
y
Computing partial derivatives is no more complicated than computing ordinary deriva-
tives.
26-Jul-2013 160
School of Mathematical Sciences Monash University
Example 29.1
If f(x, y) = sin(x) cos(y) then
f
x
=
sin(x) cos(y)
x
= cos(y)
sin(x)
x
= cos(y) cos(x)
Example 29.2
If g(x, y, z) = e
x
2
y
2
z
2
then
g
z
=
e
x
2
y
2
z
2
z
= e
x
2
y
2
z
2 (x
2
y
2
z
2
)
z
= 2ze
x
2
y
2
z
2
29.2 Higher derivatives
The result of a partial derivative is another function of one or more variables. We are thus
at liberty to take another derivative, generating yet another function. Clearly we can
repeat this any number of times (though possibly subject to some technical limitations
as noted below, see Exceptions).
Example 29.3
Let f(x, y) = sin(x) sin(y). Then we can dene g(x, y) = f/x and h(x, y) = g/x.
That is
g(x, y) =
f
x
=

_
sin(x) sin(y)
_
x
= cos(x) sin(y)
and
h(x, y) =
g
x
=

_
cos(x) sin(y)
_
x
= sin(x) sin(y)
Example 29.4
Continuing from the previous example, compute g/y.
26-Jul-2013 161
School of Mathematical Sciences Monash University
29.3 Notation
From the above example we see that h(x, y) was computed as follows
h(x, y) =
g
x
=

x
_
f
x
_
This is often written as
h(x, y) =

2
f
x
2
Now consider the case where we compute h(x, y) by rst taking a partial derivative in
x then followed by a partial derivative in y, that is
h(x, y) =
g
y
=

y
_
f
x
_
and this is normally written as
h(x, y) =

2
f
yx
Note the order on the bottom line you should read this from right to left. It tells you
that to take a partial derivative in x then a partial derivative in y.
Its now a short leap to cases where we might take say 5 partial derivatives, such as
P(x, y) =

5
Q
xyyxx
Partial derivatives that involve one or more of the independent variables are known as
mixed partial derivatives.
Example 29.5
Given f(x, y) = 3x
2
+ 2xy compute
2
f/xy and
2
f/yx. Notice anything?
Order of partial derivatives does not matter
If f is a twice-dierentiable function, then the order in which its mixed partial
derivatives are calculated does not matter. Each ordering will yield the same func-
tion. For a function of two variables this means

2
f
xy
=

2
f
yx
This is not immediately obvious but it can be proved (its a theorem!) and it is a very
useful result.
26-Jul-2013 162
School of Mathematical Sciences Monash University
Example 29.6
Use the above theorem to show that
P(x, y) =

5
Q
xyyxx
=

5
Q
yyxxx
=

5
Q
xxxyy
This allows us to simplify our notation, all we need do is record how many of each type
of partial derivative are required, thus the above can be written as
P(x, y) =

5
Q
x
3
y
2
=

5
Q
y
2
x
3
29.4 Exceptions : when derivatives do not exist
In earlier lectures we noted that at the very least a function must be continuous if it is
to have a meaningful derivative. When we take successive derivatives we may need to
revisit the question of continuity for each new function that we create.
If a function fails to be continuous at some point then we most certainly can not take
its derivative at that point.
Example 29.7
Consider the function
f(x) =
_
0 < x < 0
3x
2
0 < x <
It is easy to see that something interesting might happen at x = 0. Its also not hard to
see that the function is continuous over its whole domain, and thus we can compute its
derivative everywhere, leading to
df(x)
dx
=
_
0 < x < 0
6x 0 < x <
This too is continuous and we thus attempt to compute its derivative,
d
2
f(x)
dx
2
=
_
0 < x < 0
6 0 < x <
Now we notice that this second derivative is not continuous at x = 0. We thus can not
take any more derivatives at x = 0. Our chain of dierentiation has come to an end.
We began with a continuous function f(x) and we were able to compute only its rst two
derivatives over the domain x R. We say such that the function is twice dierentiable
26-Jul-2013 163
School of Mathematical Sciences Monash University
over R. This is also often abbreviated by saying f is C
2
over R. The symbol C reminds us
that we are talking about continuity and the superscript 2 tells us how many derivatives
we can apply before we encounter a non-continuous function. The clause over R just
reminds us that the domain of the function is the set of real numbers (, ).
We should always keep in mind that a function may only posses a nite number of
derivatives before we encounter a discontinuity. The tell-tale signs to watch out for are
sharp edges, holes or singularities in the graph of the function.
The Ten Commandments for Students of Mathematics
1. Thou shalt read Thy problem.
2. Whatsoever Thou doest to one side of ye equation, Do ye also to the other.
3. Thou must use Thy Common Sense, else Thou wilt have agpoles 9,000 metres
in height, yea ... even fathers younger than sons.
4. Thou shalt ignore the teachings of false prophets to do work in Thy head.
5. When Thou knowest not, Thou shalt look it up, and if Thy search still elude Thee,
Then Thou shalt ask the all-knowing teacher.
6. Thou shalt master each step before putting Thy heavy foot down on the next.
7. Thy correct answer does not prove that Thou hast worked Thy problem correctly.
This argument convincest none, least of all, Thy teacher.
8. Thou shalt rst see that Thou hast copied Thy problem correctly before bearing
false witness that the answer book lieth.
9. Thou shalt look back even unto Thy youth and remember Thy arithmetic.
10. Thou shalt learn, speak, write, and listen correctly in the language of mathematics,
and verily HDs and Ds shall follow Thee even unto graduation.
26-Jul-2013 164
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
30. Chain Rule, Gradient and Directional
derivatives
School of Mathematical Sciences Monash University
30.1 The Chain Rule
In a previous lecture we saw how could compute (partial) derivatives of functions of
several variables. The trick we employed was to reduce the number of independent
variables to just one (which we did by keeping all but one variable constant). There is
another way in which can achieve this reduction.
Consider a function of two variables f(x, y) and lets suppose we are given a smooth
curve in the xy-plane. Each point on this curve can be characterised by its distance
from some arbitrary starting point on the curve. In this way we can imagine that the
(x, y) pairs on this curve are given as functions of one variable, lets call it s. That is,
our curve is described by the parametric equations
x = x(s), y = y(s)
for some functions x(s) and y(s). The values of the function f(x, y) on this curve are
therefore given by
f = f(x(s), y(s))
and this is just a function of one variable s. Thus we can compute its derivative df/ds.
We will soon see that df/ds can be computed in terms of the partial derivatives.
Example 30.1
Given the curve
x(s) = 2s, y(s) = 4s
2
1 < s < 1
and the function
f(x, y) = 5x 7y + 2
compute df/ds at s = 0.
Example 30.2
Show that for the curve x(s) = s, y(s) = 2 we get df/ds = f/x.
Example 30.3
Show that for the curve x(s) = 1, y(s) = s we get df/ds = f/y.
The last two examples show that df/ds is somehow tied to the partial derivatives of f.
The exact link will be made clear in a short while (patience!).
What meaning can we assign to this number df/ds? It helps to imagine that we have
drawn a graph of f(x, y) (i.e. as a surface over the xy-plane).
Now draw the curve (x(s), y(s)) in the xy-plane and imagine walking along that curve,
lets call it C. At each point on C, f(s) is the height of the surface above the xy-plane.
If you walk a short distance s then the height might change by an amount f. The
rate at which the height changes with respect to the distance travelled is then f/s.
In the limit of innitesimal distances we recover df/ds. Thus we can interpret df/ds as
26-Jul-2013 166
School of Mathematical Sciences Monash University
measuring the rate of change of f along the curve. This is exactly what we would have
expected after all derivatives measure rates-of-change.
The rst example above showed how you could compute df/ds by rst reducing f to an
explicit function of s. It was also hinted that it is also possible to evaluate df/ds using
partial derivatives. Lets now put some paint on the canvas!
Lets go back to basics. The derivative df/ds could be calculated as
df
ds
= lim
s0
f(x(s + s), y(s + s)) f(x(s), y(s))
s
We will re-write this by adding and subtracting f(x(s), y(s +s)) just before the minus
sign. After a little rearranging we get
df
ds
= lim
s0
f(x(s + s), y(s + s)) f(x(s), y(s + s))
s
+ lim
s0
f(x(s), y(s + s)) f(x(s), y(s))
s
Now lets look at the rst limit. If we introduce x = x(s + s) x(s) then we can
write
lim
s0
f(x(s + s), y(s + s)) f(x(s), y(s + s))
s
= lim
s0
f(x(s + s), y(s + s)) f(x(s), y(s + s))
x
x
s
=
f
x
dx
ds
We can write a similar equation for the second limit. Combining the two leads us to
df
ds
=
f
x
dx
ds
+
f
y
dy
ds
This is an extremely useful and important result. It is an example of what is known as
the chain rule for functions of several variables.
The Chain Rule
Let f = f(x, y) be a dierentiable function. The chain rule for derivatives of f
along a path x = x(s), y = y(s) is
df
ds
=
f
x
dx
ds
+
f
y
dy
ds
Now that we have a head of steam its rather easy to uncover an important extension
of the above result. Suppose the path was obtained by holding some other parameter
constant. That is, imagine that the path x = x(s), y = y(s) arose from some more
26-Jul-2013 167
School of Mathematical Sciences Monash University
complicated expressions such as x = x(s, t), y = y(s, t) with t held constant. How would
our formula for the chain rule change? Not much other than we would have to keep in
mind throughout that t is constant. We encountered this issue once before and that led
to partial rather than ordinary derivatives. Clearly the same change of notation applies
here, and thus we would write
f
s
=
f
x
x
s
+
f
y
y
s
as the rst partial derivative of f with respect to s.
Lets take stock. We are given a function of two variables f = f(x, y) and we are also
given two other functions, also of two variables, x = x(s, t), y = y(s, t). Then f/s can
be calculated using the above chain rule.
Of course you could also compute f/s directly by substituting x = x(s, t), y = y(s, t)
into f(x, y) before taking the partial derivatives. Both approaches will give you exactly
the same answer.
Note that there is nothing special in the choice of symbols, x, y, s or t. You will often
nd (u, v) used rather than (s, t).
Example 30.4
Given f = f(x, y) and x = 2s + 3t, y = s 2t compute f/t directly and by way of
the chain rule.
The Chain Rule : Episode 2
Let f = f(x, y) be a dierentiable function. If x = x(u, v), y = y(u, v) then
f
u
=
f
x
x
u
+
f
y
y
u
f
v
=
f
x
x
v
+
f
y
y
v
30.2 Gradient and Directional Derivative
Given any dierentiable function of several variables we can compute each of its rst
partial derivatives. Lets do something out of the square. We will assemble these partial
derivatives as a vector which we will denote by f. So for a function f(x, y) of two
variables we dene
f =
f
x
i

+
f
y
j

The is known as the gradient of f and is often pronounced grad f.


26-Jul-2013 168
School of Mathematical Sciences Monash University
This may be pretty but what use is it? If we look back at the formula for the chain rule
we see that we can write it out as a vector dot-product,
df
ds
=
f
x
dx
ds
+
f
y
dy
ds
=
_
f
x
i

+
f
y
j

_
dx
ds
i

+
dy
ds
j

_
= (f)
_
dx
ds
i

+
dy
ds
j

_
What do we make of the vector on the far right of this equation? Its not hard to see
that it is a tangent vector to the curve (x(s), y(s)). And if we chose the parameter s to
be distance along the curve then we also see that its a unit vector.
Example 30.5
Prove the last pair of statements, that the vector is a tangent vector and that its a unit
vector.
It is customary to denote the tangent vector by t

(some people prefer u

). With the
above denitions we can now re-write the equation for a directional derivative as follows
df
ds
= t

f
Isnt that neat? The number that we calculate in this process df/ds is known as the
directional derivative of f in the direction t

.
Yet another variation on the notation is to include the tangent vector as subscript on
. Thus we also have
df
ds
=
t

f
Directional derivative
The directional derivative df/ds of a function f in the direction t

is given by
df
ds
= t

f =
t

f
where the gradient f is dened by
f =
f
x
i

+
f
y
j

and t

is a unit vector, t

= 1.
26-Jul-2013 169
School of Mathematical Sciences Monash University
Example 30.6
Given f(x, y) = sin(x) cos(y) compute the directional derivative of f in the direction
t

= ( i

+ j

)/

2.
Example 30.7
Given f = 2xi

+2yj

and x(s) = s cos(0.1), y(s) = s sin(0.1) compute df/ds at s = 1.


Example 30.8
Given f(x, y) = (xy)
2
and the vector v

= 2 i

+7j

compute the directional derivative at


(1, 1). Hint : Is v

a unit vector?
We began this discussion by restricting a function of many variables to be a function of
one variable. We achieved this by choosing a path such as x = x(s), y = y(s). We might
ask if the value of df/ds depends on the choice of the path? That is we could imagine
many dierent paths all sharing the one point, call it P, in common. Amongst these
dierent paths might we get dierent answers for df/ds?
This is a very good question. To answer it lets look at the directional derivative in the
form
df
ds
= t

f
First we note that f depends only on the values of (x, y) at P. It knows nothing about
the curves passing through P. That information is contained solely in the vector t

.
Thus if a family of curves passing through P share the same t

then we most certainly


will get the same value for df/ds for each member of that family. But what class of
curves share the same t

at P? Clearly they are all tangent to each other at P. None of


the curves cross any other curve at P.
At this point we can dispense with the curves and retain just the tangent vector t

at
P. All that we require to compute df/ds is the direction we wish to head in, t

, and the
gradient vector, f, at P. Choose a dierent t

and you will get a dierent answer for


df/ds. In each case df/ds measures how rapidly f is changing the direction of t

.
A biologist, a statistician and a mathematician are on a photo-safari in africa. They
drive out on the savannah in their jeep, stop and scout the horizon with their binoculars.
The biologist : Look! Theres a herd of zebras! And there, in the middle : A white zebra!
Its fantastic ! There are white zebras ! Well be famous !
The statistician : Its not signicant. We only know theres one white zebra.
The mathematician : Actually, we only know there exists a zebra, which is white on one
side.
26-Jul-2013 170
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
31. Tangent planes and linear approximations
School of Mathematical Sciences Monash University
31.1 Tangent planes
For functions of one variable we found that a tangent line provides a useful means of
approximating the function. It is natural to ask how we might generalise this idea to
functions of several variables.
Constructing a tangent line for a function of a single variable, f = f(x), is quite simple.
Lets just remind ourselves how we might do this. First we compute the functions value
f and its gradient df/dx at some chosen point. We then construct a straight line with
these values at the chosen point.
Example 31.1
Construct the tangent line to f = sin(x) at x = /4.
Notice that the tangent line is a linear function. Not surprisingly, for functions of several
variables we will be constructing a linear function which shares particular properties with
the original function, in particular the functions value and gradient at the chosen point.
Lets be specic. Suppose we have a function f = f(x, y) of two variables and suppose
we choose some point, say x = a, y = b. Lets call this point P. At P we can evaluate f
and all the rst partial derivatives, f/x and f/y. Now we want to construct a new
function, call it

f =

f(x, y), that shares these some numbers at P. What conditions,
apart from being linear, do we want to impose on

f? Clearly we require

f
p
= f
p
,
_


f
x
_
p
=
_
f
x
_
p
,
_


f
y
_
p
=
_
f
y
_
p
The subscript P is to remind us to impose these conditions at the point P.
As we want

f to be a linear function we could propose a function of the form

f(x, y) = C + Ax + By
We would need to carefully choose the numbers A, B, C so that we meet the above
conditions. However, it is easier (and mathematically equivalent) to choose

f(x, y) = C + A(x a) + B(y b)


In this form we nd
C = f
p
, A =
_
f
x
_
p
, B =
_
f
y
_
p
and thus we have

f(x, y) = f
p
+ (x a)
_
f
x
_
p
+ (y b)
_
f
y
_
p
This describes the tangent plane to the function f = f(x, y) at the point (a, b).
26-Jul-2013 172
School of Mathematical Sciences Monash University
Example 31.2
Prove that A, B, C are as stated.
In terms of f we can write the tangent plane in the following form

f(r

) = f
p
+ (r

p
) (f)
p
where r

= xi

+yj

. This is a nice compact formula and it makes the transition to more


variables (x, y, z ) trivial.
Example 31.3
Compute the tangent plane to the function f(x, y) = sin(x) sin(y) at (/4, /4).
The Tangent Plane
Let f = f(x, y) be a dierentiable function. The tangent plane to f at the point P
is given by

f(x, y) = f
p
+ (x a)
_
f
x
_
p
+ (y b)
_
f
y
_
p
The tangent plane may be used to approximate f at points close to P.
31.2 Linear Approximations
We have done the hard work now its time to enjoy the fruits of our labour. We can
use the tangent plane as a way to estimate the original function in a region close to the
chosen point. This is very similar to how we used a tangent line in approximations for
functions of one variable.
Example 31.4
Use the result of the previous example to estimate sin(x) sin(y) at (5/16, 5/16).
Example 31.5
Would it make sense to use the same tangent plane as in the previous example to estimate
f(5, 4)?
The bright and curious might now ask two very interesting questions, how large is the
error in the approximation and how can we build better approximations?
The answers to these questions takes us far beyond this subject but here is a very rough
guide. Suppose you are estimating f at some point a distance away from P (that is,

2
= (x a)
2
+ (y b)
2
). Then the error, |f(x, y)

f(x, y)| will be proportional to

2
. The proportionality factor will depend on the second derivatives of f (after all this
is what we left out in building the tangent plane). The upshot is that the error grows
26-Jul-2013 173
School of Mathematical Sciences Monash University
quickly as you move away from P but also, each time you halve the distance from P you
will reduce the error by a factor of four.
The answer to the second question, are there better approximations than a tangent
plane, is most certainly yes. The key idea is to force the approximation to match higher
derivatives of the original function. This leads to higher order polynomials in x and y.
Such constructions are known as Taylors series in many variables. We will revisit this
later in the course but only in the context of functions of a single variable.
26-Jul-2013 174
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
32. Maxima and minima
School of Mathematical Sciences Monash University
32.1 Maxima and minima
Suppose you run a commercial business and that by some means you have formulated
the following formula for the prot of one of your lines of business
f = f(x, y) = 4 x
2
y
2
Clearly the prot f depends on two variables x and y. Sound business practice suggest
that you would like to maximise your prots. In mathematical terms this means nd the
values of (x, y) such that f is a maximum. A simple plot of the graph of f shows us that
the maximum occurs at (0, 0). For other functions we might not be so lucky and thus
we need some systematic way of computing the points (x, y) at which f is maximised.
You would have met (in previous years) similar problems for the case of a function of
one variable. And form that you may expect that for the present problem we will be
making a statement about the derivatives of f in order that we have a maximum (i.e.
that the derivatives should be zero). Lets make this precise.
Lets denote the (as yet unknown) point at which the function is a maximum by P.
Now if we have a maximum at this point then moving in any direction from this point
should see the function decrease. That is the directional derivative must be non-positive
in every direction from P, thus we must have
df
ds
= t

(f)
p
0
for every choice of t

. Lets be tricky. Lets assume (for the moment) that (f)


p
= 0
then we should be able to compute > 0 so that t

= (f)
p
is a unit vector. If you
now substitute this into the above you will nd
(f)
p
(f)
p
0
Look carefully at the left hand side. Each term is positive (remember a

is the squared
length of a vector a

) yet the right hand side is either zero or negative. Thus this equation
does not make sense and we have to reject our only assumption, that (f)
p
= 0.
We have thus found that if f is to have a maximum at P then we must have
0 = (f)
p
This is a vector equation and thus each component of f is zero at P, that is
0 =
f
x
, and 0 =
f
y
at P
It is from these equations that we would compute the (x, y) coordinates of P.
Of course we could have posed the related question of nding the points at which a
function is minimised. The mathematics would be much the same save for a change in
words (maximum to minimum) and a corresponding change in signs. The end result
is the same, the gradient f must vanish at P.
26-Jul-2013 176
School of Mathematical Sciences Monash University
Example 32.1
Find the points at which f = 4 x
2
y
2
attains its maximum.
32.2 Local extrema
When we solve the equations
0 = (f)
p
we might get more than one point P. What do we make of these points? Some of them
might correspond to minimums while others might correspond to maximums of f. Does
this exhaust all possibilities? No, there maybe some points which can not be classied
as either a minima or a maxima of f. The three options are shown in the following
graphs.
A typical local minimum
A typical local maximum
26-Jul-2013 177
School of Mathematical Sciences Monash University
A typical saddle point
A typical case might consist of any number of points like the above. It is for this reason
that each point is referred to as a local maxima or a local minima.
32.3 Notation
Rather than continually having to qualify the point as corresponding to a minimum,
maximum or a saddle point of f we commonly lump these into the one term local
extrema.
Note when we talk of minima, maxima and extrema we are talking about the (x, y)
points at which the function has a local minimum, maximum or extremum respectively.
32.4 Maxima, Minima or Saddle point?
You may recall that for a function of one variable, f = f(x), that its extrema could be
characterised simply be evaluating the sign of the second derivative. There is a similar
test that we can apply for functions of two variables that is summarised in the following
box. Note that this result is not examinable. It is included here to whet your appetite
for the exciting things that await in your later studies in maths (you will be doing more
wont you?).
26-Jul-2013 178
School of Mathematical Sciences Monash University
What extrema was that?
If 0 = f at a point P then, at P, compute
D =

2
f
x
2

2
f
y
2

_

2
f
xy
_
2
then we have the following classication for P
A local minima when D 0 and

2
f
x
2
> 0
A local maxima when D 0 and

2
f
x
2
< 0
A Saddle point when D < 0
26-Jul-2013 179
School of Mathematical Sciences Monash University
Laborartory class exercises
Modern Engineering Mathematics, 4th ed.
Glyn James
Topic Exercises Questions
Vectors, Lines & Planes 4.2.8 17-20,23,25
4.2.10 31-34
4.3.3 52-55,59,60,62,63
Linear algebra 5.2.3 1,6,7
5.2.5 11,12,16
5.2.7 22
Matrices, Determinants & Matrix inverses 5.4.1 58,59
4.2.12 43-45
5.3.1 34,35,44
Eigenvalues & Eigenvectors 5.7.3 96,97
5.7.5 98-100,102
5.7.8 105
Hyperbolic Functions 2.7.6 82,84
8.3.13 37,38
Integration by parts 8.8.4 105-107
Improper integrals 9.2.3 1
Sequences & Series 7.2.3 1,2,4,5,12,13
7.3.4 19,21,22,24
7.6.4 41,44
9.4.4 8-17
Introduction to ODEs 10.3.6 1,2
10.4.5 3-5
1st Order ODEs 10.5.4 11,13,15,17
10.5.6 18,20
10.5.11 31-35
2nd Order homogenous ODEs 10.9.2 55-61
2nd Order inhomogenous ODEs 10.9.4 62-65
Multivariable Calculus 9.6.4 37-46
9.6.6 47,48,50-55
9.6.8 56-64
9.6.10 65-72
Maxima & Minima 9.7.3 76,78
26-Jul-2013 180
School of Mathematical Sciences Monash University
Supplementary exercises
The questions on the following pages contain no new material not already covered by
the exercies in James. They are intended for students who want to practice their craft
as far as they can (i.e., to do as many questions as possible, well done). These questions
may also be helpful for students who do not have a copy of James to hand (but make
no mistake: James is an essential book for this unit, you should obtain a copy or least
know where to nd copies in the library).
Please note that the exercises provided by James for Improper Integrals is rather thin
(just one question). So you are encouraged to complete the supplementary exercises on
Improper Integrals. This will be sucient study for questions of this kind should they
appear on the nal exam.
26-Jul-2013 181
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory Class 1
Vectors, dot product, cross product
1. Find all the vectors whose tips and tails are among the three points with coordinates
(2, 2, 3), (3, 2, 1) and (0, 1, 4).
2. Let v

= (3, 2, 2). How long is 2v

. Find a unit vector (a vector of length 1) in the


direction of v

.
3. For each pair of vectors given below, calculate the vector dot product and the angle
between the vectors.
(a) v

= (3, 2, 2) and w

= (1, 2, 1)
(b) v

= (0, 1, 4) and w

= (4, 2, 2)
(c) v

= (2, 0, 2) and w

= (3, 2, 0)
4. Given the two vectors v

= (cos(), sin(), 0) and w

= (cos(), sin(), 0), use the dot


product to derive the trigonometric identity
cos( ) = cos() cos() + sin() sin().
5. Use the dot product to determine which of the following two vectors are perpendicular
to one another: u

= (3, 2, 2), v

= (1, 2, 2), w

= (2, 1, 2).
6. For each pair of vectors given below, calculate the vector cross product. Assuming that
the vectors dene a parallelogram, calculate the area of the parallelogram.
(a) v

= (3, 2, 2), w

= (1, 2, 1)
(b) v

= (0, 1, 4), w

= (4, 2, 2)
(c) v

= (2, 0, 2), w

= (3, 2, 0)
7. Calculate the volume of the parallelepiped dened by the three vectors u

= (3, 2, 2), v

=
(1, 2, 2), w

= (2, 1, 2).
8. Verify that v

= w

.
School of Mathematical Sciences Monash University
Lines and planes
9. Consider the points (1, 2, 1) and (2, 0, 3).
(a) Find a vector equation of the line through these points in parametric form.
(b) Find the distance between this line and the point (1, 0, 1). (Hint: Use the para-
metric form of the equation and the dot product.)
10. Find an equation of the plane that passes through the points (1, 2, 1), (2, 0, 1) and
(1, 1, 0).
11. Consider a plane dened by the equation 3x + 4y z = 2 and a line dened by the
following vector equation (in parametric form)
x(t) = 2 2t, y(t) = 1 + 3t, z(t) = t.
(a) Find the point where the line intersects the plane. (Hint: Substitute the parametric
form into the equation of the plane.)
(b) Find a normal vector to the plane.
(c) Find the angle at which the line intersects the plane. (Hint: Use the dot product.)
12. Find the distance between the parallel planes dened by the equations 2xy +3z = 4
and 2x y + 3z = 24. (Hint: Use the cross product to construct a line normal to both
planes, then use problem 11.)
13. Consider two planes dened by the equations 3x + 4y z = 2 and 2x + y + 2z = 6.
(a) Find where the planes intersect the x, y and z axes.
(b) Find normal vectors for the planes.
(c) Find an equation of the line dened by the intersection of these planes. (Hint: Use
the normal vectors to dene the direction of the line.)
(d) Find the angle between these two planes.
14. Find the minimum distance between the two lines dened by
x(t) = 1 + t, y(t) = 1 3t, z(t) = 2 + 2t
and
x(s) = 3s, y(s) = 1 2s, z(s) = 2 s
(Hint: Use scalar projection as demonstrated in the lecture notes. Alternatively, dene
the lines within parallel planes and then go back to problem 12.)
26-Jul-2013 183
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory Class 1 Solutions
Vectors, dot product, cross product
1. (0, 0, 0) (1, 4, 2) (2, 1, 7) (3, 3, 5)
2. | 2v

| = 2

17 ,
v
| v

|
=
1

17
(3, 2, 2)
3. (a) v

= 1, = arccos
_
1

617
_
1.4716 radians
(b) v

= 10, = arccos
_
10

1724
_
2.0887 radians
(c) v

= 6, = arccos
_
6

813
_
2.1998 radians
4. v

= |v

||w

| cos( ) = 1 1 cos( ) = cos() cos() + sin() sin()


5. u

and w

6. (a) v

= (6, 1, 8) |v

| =

101
(b) v

= (6, 16, 4) |v

| = 2

77
(c) v

= (4, 6, 4) |v

| = 2

17
7. (u

) w

= 4
8. Yes, it is correct!
Lines and planes
9. (a) x(t) = 1 + t, y(t) = 2 2t, z(t) = 1 + 4t
(b)
2
7

14
10. 2x + y + 7z = 3
11. (a) (2, 1, 0)
(b) (3, 4, 1)
(c)

2
arccos
_

91
26
_
0.37567 radians
School of Mathematical Sciences Monash University
12.

56
13. (a) (2/3, 0, 0), (0, 1/2, 0), (0, 0, 2) and (3, 0, 0), (0, 6, 0), (0, 0, 3)
(b) (3, 4, 1) and (2, 1, 2)
(c) x(t) = 2 + 9t, y(t) = 2 4t, z(t) = 11t
(d) arccos
_
2
39

26
_
1.835 radians
14.

3
26-Jul-2013 185
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory class 2
Row operations and linear systems
Solve each of the following system of equations using Gaussian elimination with back-
substitution. Be sure to record the details of each row-operation (for example, as a note
on each row of the form (2) 2(2) 3(1).)
1.
J + M = 75
J 4M = 0
2.
x + y = 5
2x + 3y = 1
3.
x + 2y z = 6
2x + 5y z = 13
x + 3y 3z = 4
4.
x + 2y z = 6
x + 2y + 2z = 3
2x + 5y z = 13
5.
2x + 3y z = 4
x + y + 3z = 1
x + 2y z = 3
6. Repeat the last two questions, this time using Gaussian elimination (i.e. no back-
substitution).
School of Mathematical Sciences Monash University
Under-determined systems
7. Using Gaussian elimination with back-substitution to nd all possible solutions for the
following system of equations
x + 2y z = 6
x + 3y = 7
2x + 5y z = 13
8. Find all possible solutions for the system (sic) of equations
x + 2y z = 6
(Hint : You have one equation but three unknowns. You will need to introduce two free
parameters).
Matrices
9. Evaluate each of the following matrix operations
2
_
1 1
1 4
_

_
2 1
3 1
_
,
_
1 1
1 4
_ _
2 1
3 1
_
,
_
1 1 3
1 4 2
_
_
_
2 1
3 1
1 2
_
_
10. Rewrite the equations for questions 1,2 and 3 in matrix form. Hence write down the
coecient and augmented matrices for questions 1,2 and 3.
11. Repeat the row-operations part of questions 4 and 5 using matrix notation (should be
easy).
Matrix inverses
12. Compute the inverse A
1
of the following matrices
A =
_
1 1
1 4
_
A =
_
_
2 3 1
1 1 3
1 2 1
_
_
Verify that A
1
A = I and AA
1
= I.
13. Use the result of the previous question to solve the system of equations in questions 1
and 5.
26-Jul-2013 187
School of Mathematical Sciences Monash University
Matrix determinants
14. Compute the determinant for the coecient matrices in questions 7 and 8. What do
you observe?
15. For the matrix
A =
_
_
2 3 1
1 1 3
1 2 1
_
_
compute the determinant twice, rst by expanding about the top row and second by
expanding about the second column.
16. Given
A =
_
1 1
1 4
_
, B =
_
2 1
3 1
_
compute det(A), det(B) and det(AB). Verify that det(AB) = det(A) det(B).
26-Jul-2013 188
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory Class 2 Solutions
Row operations and linear systems
1. J = 60, M = 15 2. x = 14, y = 9 3. x = 7, y = 0, z = 1
4. x = 1, y = 2, z = 1 5. x = 1, y = 2, z = 0
Under-determined systems
7. Solution is x(t) = 4 + 3t, y(t) = 1 t, z(t) = t where t is a parameter, < t < .
8. Solution is x(u, v) = u 2v + 6, y(u, v) = v, z(u, v) = u where u, v are parameters,
< u, v < .
Matrices
9. Solutions are,
_
0 3
1 9
_ _
5 0
10 5
_ _
8 6
8 1
_
10. Coecient and augmented matrices are
Q1.
_
1 1
1 4
_
,
_
1 1 75
1 4 0
_
Q2.
_
1 1
2 3
_
,
_
1 1 5
2 3 1
_
Q3.
_
_
1 2 1
2 5 1
1 3 3
_
_
,
_
_
1 2 1 6
2 5 1 13
1 3 3 4
_
_
School of Mathematical Sciences Monash University
Matrix inverses
12. Inverses are,
A
1
=
1
5
_
4 1
1 1
_
A
1
=
1
3
_
_
7 1 10
4 1 7
1 1 1
_
_
Matrix determinants
14. First add rows of zeroes to make the coecient matrices square. Then compute the
determinants, both are zero. This tells you that the system is under-determined and
that you will need to introduce parameters during the back-substitution.
15. Determinant = 3.
16. det(A) = 5, det(B) = 5 and det(AB) = 25.
26-Jul-2013 190
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory class 3
Matrices and Determinants Pt 2.
1. Compute the following determinants using expansions about any suitable row or col-
umn.
(a)

1 2 3
3 2 2
0 9 8

(b)

4 3 2
1 7 8
3 9 3

(c)

1 2 3 2
1 3 2 3
4 0 5 0
1 2 1 2

(d)

1 5 1 3
2 1 7 5
1 2 1 0
3 1 0 1

2. Recompute the determinants in the previous question this time using row operations
(ie., Gaussian elimination).
3. Which of the following statements are true? Which are false?
(a) If A is a 33 matrix with a zero determinant, then one row of A must be a multiple
of some other row.
(b) Even if any two rows of a square matrix are equal, the determinant of that matrix
may be non-zero.
(c) If any two columns of a square matrix are equal then the determinant of that matrix
is zero.
(d) For any pair of n n matrices, A and B, we always have det(A + B) = det(A) +
det(B)
(e) Let A be an 3 3 matrix. Then det(7A) = 7
3
det(A).
(f) If A
1
exists, then det(A
1
) = det(A).
4. Given
A =
_
1 k
0 1
_
Compute A
2
, A
3
and hence write down A
n
for n > 1.
School of Mathematical Sciences Monash University
5. Assume that A is square matrix with an inverse A
1
. Prove that det(A
1
) = 1/ det(A)
6. Let
A =
_
5 2
2 1
_
Show that
A
2
6A + I = 0
where I is the 2 2 identity matrix. Use this result to compute A
1
.
7. Consider the following pair of matrices
A =
_
_
11 18 7
a 6 3
3 5 2
_
_
, B =
_
_
3 1 12
b 1 5
2 1 6
_
_
Compute the values of a and b so that A is the inverse of B while B is the inverse of A.
8. Here is a 2 2 matrix equation
_
a b
c d
_
=
_
e f
g h
_ _
p q
r s
_
Show that this is equivalent to the following sets of equations
_
a
c
_
= p
_
e
g
_
+ r
_
f
h
_
and
_
b
d
_
= q
_
e
g
_
+ s
_
f
h
_
9. Use the result of the previous question to show that if the original 22 matrix equation
is written as
A = EP
then the columns of A are linear combinations of the columns of E.
10. Following on from the previous two questions, show that the rows of A can be written
as linear combinations of the rows of P.
26-Jul-2013 192
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory class 3 Solutions
Matrices and Determinants Pt 2.
1. If possible, use a row or column that contains one or more zeros.
(a) 31 =

1 2 3
3 2 2
0 9 8

(b) 165 =

4 3 2
1 7 8
3 9 3

(c) 0 =

1 2 3 2
1 3 2 3
4 0 5 0
1 2 1 2

(d) 162 =

1 5 1 3
2 1 7 5
1 2 1 0
3 1 0 1

3. Which of the following statements are true? Which are false?


(a) False (b) False (c) True
(d) False (e) True (f) False
4. Compute A
2
and A
3
and note the pattern.
A
n
=
_
1 nk
0 1
_
6.
A
1
= 6I A =
_
1 2
2 5
_
7. Require that AB = I and BA = I. Then a = 4 and b = 1.
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory class 4
Matrix operations
1. Suppose you are given a matrix of the form
R() =
_
cos sin
sin cos
_
Consider now the unit vector v

= [1, 0]
T
in a two dimensional plane. Compute R()v

.
Repeat your computations this time using w

= [0, 1]
T
. What do you observe? Try
thinking in terms of pictures, look at the pair of vectors before and after the action of
R().
2. You may have recognised the two vectors in the previous question to be the familar basis
vectors for a two dimensional space, i.e., i

and j

. We can express any vector as a linear


combination of i

and j

, that is
u

= a i

+ bj

for some numbers a and b. Given what you learnt from the previous question, what do
you think will be result of R()u

? Your answer can be given in simple geometrical terms


(e.g., in pictures).
3. Give reasons why you expect R( + ) = R()R(). Hence deduce that
cos( + ) = cos cos sin sin
sin( + ) = sin cos + sin cos
4. Give reasons why you expect R()R() = R()R(). Hence prove that the rotation
matrices R() and R() commute.
5. Show that det R() = +1.
6. Given the above form for R() write down, without doing any computations, the inverse
of R().
School of Mathematical Sciences Monash University
Eigenvectors and eigenvalues
A square matrix A has an eigenvector v with eigenvalue provided
Av = v
The vector v would normally be written as a column vector. Its transpose v
T
is a row
vector.
The eigenvalues are found by solving the polynomial equation
0 = det(A I)
7. Compute the eigenvalues and eigenvectors of the following matrices.
(a)
_
4 2
5 3
_
(b)
_
6 1
3 2
_
(c)
_
5 3
3 1
_
8. Given that one eigenvalue is = 4, compute the remaining eigenvalues of the following
matrices.
(a)
_
_
1 3 3

2
3 1 3

2
3

2 3

2 2
_
_
(b)
_
_
3 1 3

2
1 3 3

2
3

2 3

2 2
_
_
9. Compute the eigenvectors for each matrix of the previous question. Verify that the
eigenvectors of part (b) are mutually orthogonal (i.e., 0 = v
T
1
v
2
, 0 = v
T
1
v
3
and 0 = v
T
2
v
3
).
10. Suppose the matrix A has eigenvectors v with corresponding eigenvalues . Show that
v is an eigenvector of A
n
. What is its corresponding eigenvalue?
11. If , v are an eigenvalue-eigenvector pair for A then show that v is also an eigenvector
of A.
12. Suppose the matrix A has eigenvectors v with corresponding eigenvalues . Deduce the
eigenvectors and eigenvalues of R
1
AR where R is a non-singular matrix.
13. Let A be any matrix of any shape. Show that A
T
A is a symmetric square matrix.
26-Jul-2013 195
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory class 4 Solutions
Matrix operations
1. Each of the vectors will have been rotated about the origin by the angle in a counter-
clockwise direction.
2. The rotation observed in the previous question also applies to the general vector u

. Thus
R() is often referred to as a rotation matrix. Matrices like this (and their 3 dimensional
counterparts) are used extensivly in computer graphics.
3. Any object rotated rst by and then by could equally have been subject to a single
rotation by +. The resulting objects must be identical. Hence R( +) = R()R().
4. Regardless of the order in which the rotations have been applied the nett rotation will
be the same. Thus R()R() = R()R(). Equally, you could have started by writing
+ = + , then R( + ) = R( + ) and so R()R() = R()R().
5.
det R() =

cos sin
sin cos

= 1
6. The inverse of R() is R().
Eigenvectors and eigenvalues
7.
(a) = 1 and 2 (b) = 3 and 5 (c) = 2 (a double root)
8.
(a) = 8 and 4 (a double root) (b) = 8, 4 and 4
9. In part (a) there is a double root = 4. In this case there are two linearly independent
eigenvectors. Your may answers may appear dierent from those given here, you will
need to check that your eigenvectors are linear combinations of those given here. Also,
remember that any scaling is allowed for an eigenvector.
(a) = 8 v = (1, 1,

2)
T
= 4 v = (2, 0,

2)
T
= 4 v = (1, 1, 0)
T
(b) = 8 v = (1, 1,

2)
T
= 4 v = (1, 1, 0)
T
= 4 v = (1, 1,

2)
T
School of Mathematical Sciences Monash University
10. The eigenvalue of A
n
will be
n
.
11. This is trivial, just multiply the eigenvalue equation Av = v by .
12. The matrix R
1
AR will have as an eigenvalue with eigenvector R
1
v.
13. Use (PQ)
T
= Q
T
P
T
and (A
T
)
T
= A to show that (A
T
A)
T
= A
T
A. Hence A
T
A is
symmetric.
26-Jul-2013 197
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory class 5
Integration by parts
1. Evaluate each of the following using integration by parts. Recall that
_
f
dg
dx
dx = fg
_
g
df
dx
dx
(a)
_
x cos(x) dx (b)
_
xe
x
dx
(c)
_
y
_
y + 1 dy (d)
_
x
2
log(x) dx
(e)
_
sin
2
() d (f)
_
cos
2
() d
(g)
_
sin() cos() d (h)
_
sin
2
() d
2. Use integration by parts twice to nd
_
e
x
sin(x) dx and
_
e
x
cos(x) dx.
3. Use a substitution and an integration by parts to evaluate each of the following
(a)
_
(3x 7) sin(5x + 2) dx (b)
_
cos(x) sin(x)e
cos(x)
dx
(c)
_
e
2x
cos (e
x
) dx (d)
_
e

x
dx
4. Spot the error in the following calculation.
We wish to compute
_
dx/x. For this we will use integration by parts with u = 1/x and
dv = dx. This gives us du = dx/x
2
and v = x. Thus using
_
u dv = uv
_
v du we
nd _
dx
x
= 1 +
_
dx
x
and thus 0 = 1. (If this answer does not cause you serious grief then a career in
accountancy beckons).
School of Mathematical Sciences Monash University
Improper integrals
5. Decide which of the following improper integrals will converge and which will diverge.
(a)
_
1
0
1
x
dx (b)
_
1
0
1
x
1/4
dx
(c)
_
1
0
1
y
4
dy (d)
_

0
e
2x
dx
(e)
_

0
1
1 +
2
d
Comparison test for Improper integrals
6. Use a suitable comparison function to decide which of the following integrals will converge
and which will diverge.
(a)
_
1
0
e
x
x
dx (b)
_
1
0
1
1 x
1/4
dx
(c)
_
1
0
e
y
y
4
dy (d)
_

0
sin
2
(x)e
2x
dx
(e)
_

0
e

1 +
2
d (f)
_
1
0
1
x(1 x
2
)
dx
26-Jul-2013 199
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory Class 5 Solutions
Integration by parts
1. (a)
_
x cos(x) dx = cos(x) + x sin(x) + C
(b)
_
xe
x
dx = e
x
xe
x
+ C
(c)
_
y
_
y + 1 dy =
2
3
y (y + 1)
3/2

4
15
(y + 1)
5/2
+ C
(d)
_
x
2
log(x) dx =
x
3
3
log(x)
x
3
9
+ C
(e)
_
sin
2
() d =
1
2
( cos() sin()) + C
(f)
_
cos
2
() d =
1
2
( + cos() sin()) + C
(g)
_
sin() cos() d =
1
2
sin
2
() + C
(h)
_
sin
2
() d =

2
cos() sin() +
1
4
sin
2
() +
1
4

2
+ C
2. (a)
_
e
x
sin(x) dx =
e
x
2
(sin(x) cos(x)) + C
(b)
_
e
x
cos(x) dx =
e
x
2
(sin(x) + cos(x)) + C
3. (a)
_
(3x 7) sin(5x + 2) dx =
3
25
sin(5x + 2) +
1
5
(7 3x) cos(5x + 2) + C
(b)
_
cos(x) sin(x)e
cos(x)
dx = e
cos(x)
(1 cos(x)) + C
School of Mathematical Sciences Monash University
(c)
_
e
2x
cos (e
x
) dx = cos(e
x
) + e
x
sin(e
x
) + C
(d)
_
e

x
dx = 2e

x
_
x 1
_
+ C
4. Did we forget an integration constant? (And so with the natural order restored, fears of
a career in accountancy fade from view.)
Improper integrals
5. Decide which of the following improper integrals will converge and which will diverge.
(a)
_
1
0
1
x
dx diverges (b)
_
1
0
1
x
1/4
dx converges to 4/3
(c)
_
1
0
1
y
4
dy diverges (d)
_

0
e
2x
dx converges to 1/2
(e)
_

0
1
1 +
2
d converges to /2 (f)
_
2
0
1
1 x
2
dx diverges
(g)
_
2
0
1
x(x + 2)
dx diverges (h)
_
2
0
1
x(x 2)
dx diverges
Comparison test for Improper integrals
6. Use a suitable comparison function to decide which of the following integrals will converge
and which will diverge.
(a)
_
1
0
e
x
x
dx diverges, use
1
x
<
e
x
x
over 0 < x < 1
(b)
_
1
0
1
1 x
1/4
dx diverges, use x < x
1/4
over 0 < x < 1
(c)
_
1
0
e
y
y
4
dy diverges, use
1
3y
4
<
e
y
y
4
over 0 < y < 1
(d)
_

0
sin
2
(x)e
2x
dx converges, use sin
2
(x)e
2x
< e
2x
over 0 < x <
(e)
_

0
e

1 +
2
d converges, use
e

1 +
2
<
1
1 +
2
over 0 < <
(f)
_
1
0
1
x(1 x
2
)
dx diverges, use
1
x
<
1
x(1 x
2
)
over 0 < x < 1
26-Jul-2013 201
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory class 6
Sequences
1. Find the limit, if it exists, for each of the following sequences
(a) 1, +
1
2
,
1
3
, +
1
4
, ,
(1)
n
n+1
,
(b)
1
2
,
2
3
,
3
4
, ,
n+1
n+2
,
(c) a
n
=
1
n+1
, n 0
(d) a
n
=
1
n+2

1
n+1
, n 0
(e) a
n
=
_

_
1 +
1
n+1
, n even
1
1
n+1
, n odd
(f) a
n
=
_

_
e
n
, n 100
e
n
, 0 n < 100
(g) a
n
= sin(
n
4
) (Hint : Write out the rst few terms.)
2. Consider the sequence dened by
a
n+1
= a
n
+
_
1
2
_
n+1
, n 0
with a
0
= 1.
(a) Write out the rst few terms a
0
, , a
4
.
(b) Can you express a
5
in terms of
1
2
a
4
?
(c) Generalize this result to express a
n+1
in terms of
1
2
a
n
.
(d) Can you express a
n
as a sum

n
k=0
b
k
for some set of b
k
?
(e) Suppose the limit lim
n
a
n
exists. Use the result of (c) to deduce the limit.
(f) Determine the values of for which the sequence a
n+1
= a
n
+
n
converges.
School of Mathematical Sciences Monash University
Series
3. Which of the following statements are true?
(a) The innite series

n=0
a
n
converges whenever lim
n
|a
n
| = 0.
(b) The harmonic series

n=0
1/(n + 1) converges.
(c) If the series

n=0
|a
n
| converges then

n=0
a
n
also converges.
(d) If

n=0
a
n
diverges then

n=0
(1)
n
a
n
converges.
(e) If lim
n
|
an+1
an
| > 1 then

n=0
a
n
converges.
The Integral Test
4. Establish the convergence (or divergence) of the following series using the integral test.
(a)

n=0
1

n+1
(b)

n=0
1
(n+1)

, > 1
(c)

n=0
1
n
2
+1
(d)

n=0
1
(n+1)(n+2)
(Hint : First establish a comparison with

n=0
(n + 1)
2
then use
the integral test.)
The Comparison Test
5. Determine the convergence or otherwise of the following series using the suggested series
for comparison.
(a)

n=0
n+2
n+1
compare with

n=0
1
(b)

n=0
1
(2+1/(n+1))
n+1
compare with

n=0
1
2
n+1
(c)

n=0
2+sin n
n+1
compare with

n=0
1
n+1
(d)

n=0
3
n
n+1
compare with

n=0
_
1
3
_
n
The Ratio Test
6. Use the ratio test to examine the convergence of the following series.
26-Jul-2013 203
School of Mathematical Sciences Monash University
(a)

n=0

n
, || > 1
(b)

n=0
x
n
n+1
, |x| < 1
(c)

n=0
n
1n
(d)

n=0
n
3
e
n+2
7. What does the ratio test tell you about the convergence of

n=0
1
(n + 1)
2
Can you establish the convergence of this series by some other method?
8. The Starship USS Enterprise is being pursued by a Klingon warship. The dilithium
crystals couldnt handle the warp speed and so it would appear that Captain Kirk and
his crew are about to become as one with the inter-galactic dust cloud.
Spock : Captain, the enemy are 10 light years away and are closing fast.
Kirk : But Spock, by the time they travel the 10 light years we will have travelled
a further 5 light years. And when they travel those 5 light years we will
have moved ahead by a further 2.5 light years, and so on forever. Spock,
they will never capture us!
Spock : I must inform the captain that he has made a serious error of logic.
What was Kirks mistake? How far will Kirks ship travel before being caught?
26-Jul-2013 204
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory class 6 Solutions
Sequences
1. (a) 0 (b) 1 (c) 0 (d) 0
(e) 1 (f) 0 (g) Limit does not exist
2. This is the geometric series. It converges for || < 1.
Series
3. (a) False (b) False (c) True (d) False
(e) False
The Integral Test
4. (a) Diverges (b) Converges (c) Converges (d) Converges
The Comparison Test
5. (a) Diverges (b) Converges (c) Diverges (d) Converges
The Ratio Test
6. (a) Converges (b) Converges (c) Converges (d) Converges
7. The series converges and this could also be established using the integral test.
8. Clearly the fast ship must catch the slow ship in a nite time. Yet Kirk has put an
argument which shows that his slow ship will still be ahead of the fast ship after each
cycle (a cycle ends when the fast ship just passes the location occupied by the slow ship
at the start of the cycle). Each cycle takes a nite amount of time. The total elapsed
time is the sum of the times for each cycle. Kirks error was to assume that the time
taken for an innite number of cycles must be innite. We know that this is wrong an
innite series may well converge to a nite number.
School of Mathematical Sciences Monash University
Given the information in the question we can see that the fast ship is initially 10 light
years behind the slow ship and that it is traveling twice as fast as the slow ship. Suppose
the fast ship is traveling at v light years per year. The distance traveled by the fast ship
decreases by a factor of 2 in each cycle. Hence the time interval for each cycle also
decreases by a factor of 2 in each cycle. The total time taken will then be
Time =
10 + 5 + 2.5 + 1.25 + ...
v
=
10
v
_
1 +
1
2
+
1
4
+
1
8

_
=
10
v
1
1
1
2
=
10
v/2
We expect that this must be time taken for the fast ship to catch the slow ship. The
fast ship is traveling at speed v while the slow ship is traveling at speed v/2. Thus the
fast ship is approaching the slow ship at a speed v/2 and it is initially 10 light years
behind. Hence it will take the Klingons 10/(v/2) light years to catch Kirks starship.
26-Jul-2013 206
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory class 7
Power series
1. Find the radius of convergence for each of the following power series
(a) f(x) =

k=0
kx
k
3
k
(b) g(x) =

k=0
x
k
3
k
k!
(c) h(x) =

k=0
k
2
x
k
(d) p(x) =

k=0
x
2k
log(1+k)
(e) q(x) =

k=0
k!(x1)
k
2
k
k
k
(f) r(x) =

k=0
(1 + k)
k
x
k
Maclaurin Series
2. Find the rst 4 non-zero terms in Maclaurin series for each of the following functions
(a) f(x) = cos(x) (b) f(x) = sin(2x)
(c) f(x) = log(1 + x) (d) f(x) =
1
1+x
2
(e) f(x) = arctan(x) (f) f(x) =

1 x
2
3. Use the previous results to obtain the rst 2 non-zero terms in the Maclaurin series for
the following functions.
(a) f(x) = cos(x) sin(2x) (c) f(x) = log(1 + x
2
)
(d) f(x) =
1
1+cos
2
(x)
(e) f(x) = arctan(arctan(x))
As the algebra in some parts of this question is rather tedious, you might like to do this
question using Scientic Notebook.
School of Mathematical Sciences Monash University
Taylor Series
4. Compute the Taylor series, about the the given point, for each of the following functions.
(a) f(x) =
1
x
, a = 1 (b) f(x) =

x, a = 1
(c) f(x) = e
x
, a = 1 (d) f(x) = log x, a = 2
5. (a) Compute the Taylor series for e
x
(b) Hence write down the Taylor series for e
x
2
(c) Use the above to obtain an innite series for the function
s(x) =
_
x
0
e
u
2
du
6. (a) Compute the Taylor series, around x = 0, for log(1 + x) and log(1 x).
(b) Hence obtain a Taylor series for f(x) = log
_
1+x
1x
_
(c) Compute the radius of convergence for the Taylor series in part (b).
(d) Show that the function dened by y(x) =
1+x
1x
has a unique inverse for almost
all values of y.
(e) Use the above results to obtain a power series for log(y) valid for 1 < |y| < .
26-Jul-2013 208
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory class 7 Solutions
Power series
1. (a) R = 3 (b) R =
(c) R = 1 (d) R = 1
(e) R = 2e, note lim
n
(1 + x/n)
n
= e
x
(f) R = 0
Maclaurin Series
2. (a) cos(x) = 1
1
2
x
2
+
1
24
x
4

1
720
x
6
+
(b) sin(2x) = 2x
4
3
x
3
+
4
15
x
5

8
315
x
7
+
(c) log(1 + x) = x
1
2
x
2
+
1
3
x
3

1
4
x
4
+
(d)
1
1+x
2
= 1 x
2
+ x
4
x
6
+
(e) arctan(x) = x
1
3
x
3
+
1
5
x
5

1
7
x
7
(f)

1 x
2
= 1
1
2
x
2

1
8
x
4

1
16
x
6
+
3. (a) cos(x) sin(2x) = 2x
7
3
x
3
+ (c) log(1 + x
2
) = x
2

1
4
x
4
+
(d)
1
1+cos
2
(x)
=
1
2
+
1
4
x
2
+ (e) arctan(arctan(x)) = x
2
3
x
3
+
Taylor Series
4. (a)
1
x
= 1 (x) + (x 1)
2
(x 1)
3
+ (x 1)
4
+
(b)

x = 1 +
1
2
(x 1)
1
8
(x 1)
2
+
1
16
(x 1)
3
+
(c) e
x
= e
1
(1 + (x + 1) +
1
2
(x + 1)
2
+
1
6
(x + 1)
3
+
(d) log
e
x = log
e
(2) +
1
2
(x 2)
1
8
(x 2)
2
+
1
24
(x 2)
3
+
School of Mathematical Sciences Monash University
5. (a) e
x
= 1 + x +
1
2
x
2
+
1
6
x
3
+
1
24
x
4
+
(b) e
x
2
= 1 x
2
+
1
2
x
4

1
6
x
6
+
1
24
x
8
+
(c) s(x) =
_
x
0
e
u
2
= x
1
3
x
3
+
1
10
x
5

1
42
x
7
+
1
216
x
9
+
6. (a) log
e
(1 + x) = x
1
2
x
2
+
1
3
x
3

1
4
x
4
+ =

n=1
(1)
(n+1)
n
x
n
log
e
(1 x) = x
1
2
x
2

1
3
x
3

1
4
x
4
+ =

n=1
1
n
x
n
(b) log
e
_
1+x
1x
_
= 2x + 2
1
3
x
3
+ 2
1
5
x
5
+ = 2

n=1
1
2n1
x
2n1
, R = 1
(c) x =
y1
y+1
, y = 1
(d) log
e
(y) = 2

n=1
1
2n1
x
2n1
, x = (y 1)/(y + 1)
26-Jul-2013 210
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory class 8
Separable rst order ODEs
1. Find the general solution for each of the following seperable ODEs
(a)
dy
dx
= 2xy (b) y
dy
dx
+ sin(x) = 0
(c) sin(x)
dy
dx
+ y cos(x) = 2 cos(x) (d)
1 + dy/dx
1 dy/dx
=
1 y/x
1 + y/x
Non-separable rst order ODEs
2. For each of the following ODEs nd any particular solution.
(a)
dy
dx
+ y = 1 (b)
dy
dx
+ 2y = 2 + 3x
(c)
dy
dx
y = e
2x
(d)
dy
dx
y = e
x
(e)
dy
dx
+ 2y = cos(2x) (f)
dy
dx
2y = 1 + 2x sin(x)
3. Find the general solution of the homogenous equation for each of the ODEs in the
previous question. Hence obtain the general solution of the ODE.
Integrating factor
4. Use an integrating factor to nd the general solution for each of the following ODEs
(a)
dy
dx
+ 2y = 2x (b)
dy
dx
+
2
x
y = 1
(c)
dy
dx
+ cos(x)y = 3 cos(x) (d) sin(x)
dy
dx
+ cos(x)y = tan(x)
Second order homogenous ODEs
5. Find the general solution for each of the following ODEs.
School of Mathematical Sciences Monash University
(a)
d
2
y
dx
2
+
dy
dx
2y = 0 (b)
d
2
y
dx
2
9y = 0
(c)
d
2
y
dx
2
+ 2
dy
dx
+ 2y = 0 (d)
d
2
y
dx
2
+ 6
dy
dx
+ 10y = 0
(e)
d
2
y
dx
2
4
dy
dx
+ 4y = 0 (f)
d
2
y
dx
2
+ 6
dy
dx
+ 9y = 0
6. Find the particular solution, for the corresponding ODE in the previous question, that
satises the following boundary conditions.
(a) y(0) = 1 and y(1) = 0 (b) y(0) = 0 and y(1) = 1
(c) y(0) = 1 and y(+/2) = +1 (d) y(0) = 1 and
dy
dx
= 0 at x = 0
(e) y(0) = 1 and
dy
dx
= 0 at x = 1 (f)
dy
dx
= 0 at x = 0 and
dy
dx
=
1 at x = 1
Second order non-homogenous ODEs
7. Find the general solution for each of the following ODEs.
(a)
d
2
y
dx
2
+
dy
dx
2y = 1 + x (b)
d
2
y
dx
2
9y = e
3x
(c)
d
2
y
dx
2
+ 2
dy
dx
+ 2y = sin(x) (d)
d
2
y
dx
2
+ 6
dy
dx
+ 10y = e
2x
cos(x)
(e)
d
2
y
dx
2
4
dy
dx
+ 4y = 2x (f)
d
2
y
dx
2
+ 6
dy
dx
+ 9y = cos(x)
26-Jul-2013 212
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory class 8 Solutions
Separable rst order ODEs
1. (a) y = Ce
x
2
(b) y =
_
2 cos(x) + C
(c) y = 2 +
C
sin(x)
(d) y =
C
x
Non-separable rst order ODEs
2. (a) y = 1 (b) y =
1
4
+
3x
2
(c) y = e
2x
(d) y = xe
x
(e) y =
1
4
cos(2x) +
1
4
sin(2x) (f) y = 1 x +
1
5
cos(x) +
2
5
sin(x)
3. (a) y = 1 + Ce
x
(b) y =
1
4
+
3x
2
+ Ce
2x
(c) y = e
2x
+ Ce
x
(d) y = xe
x
+ Ce
x
(e) y =
1
4
cos(2x) +
1
4
sin(2x) + Ce
2x
(f) y = 1 x +
1
5
cos(x) +
2
5
sin(x) +
Ce
2x
Integrating factor
4. (a) y = x
1
2
+ Ce
2x
(b) y =
x
3
+
C
x
2
(c) y = 3 + Ce
sin(x)
(d) y =
C log
e
(cos(x))
sin(x)
School of Mathematical Sciences Monash University
Second order homogenous ODEs
5. (a) y = Ae
x
+ Be
2x
(b) y = Ae
3x
+ Be
3x
(c) y = (Acos(x) + Bsin(x)) e
x
(d) y = (Acos(x) + Bsin(x)) e
3x
(e) y = (A + Bx) e
2x
(f) y = (A + Bx) e
3x
6. (a) y(x) =
1
e
3
1
_
e
32x
e
x
_
(b) y(x) =
e
3x
e
3x
e
3
e
3
(c) y(x) =
_
cos(x) + e
/2
sin(x)
_
e
x
(d) y(x) = ((3 sin(x) + cos(x)) e
3x
(e) y(x) =
_
1
2x
3
_
e
2x
(f) y(x) =
1
9
(1 + 3x) e
33x
Second order non-homogenous ODEs
7. (a) y =
3
4

x
2
+ Ae
x
+ Be
2x
(b) y =
_
A +
x
6
_
e
3x
+ Be
3x
(c) y =
1
5
(2 cos(x) + sin(x)) + (Acos(x) + Bsin(x))e
x
(d) y =
1
145
(5 cos(x) + 2 sin(x)) e
2x
+ (Acos(x) + Bsin(x))e
3x
(e) y =
1 + x
2
+ (A + Bx)e
2x
(f) y =
1
50
(4 cos(x) + 3 sin(x)) + (A + Bx)e
3x
26-Jul-2013 214
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory class 9
lHopitals rule
1. Use lHopitals rule to verify the following limits
(a) 2 = lim
x1
x
2
1
x + 1
(b)
4
5
= lim
x0
sin(4x)
sin(5x)
(c)
1

2
= lim
x1
1 x + log(x)
1 + cos(x)
(d) 0 = lim
x
log(log(x))
x
(e)
1
4
= lim
x0
x
tan
1
(4x)
(f) 0 = lim
x
e
x
log(x)
2. Prove that for any n > 0
0 = lim
x
x
n
e
x
3. Prove that for any n > 0
0 = lim
x
x
n
log(x)
Coupled rst order ODEs
4. Solve each of the following coupled ODEs by rst dierentiating each equation and
then making suitable combinations to de-couple the equations. Verify your solutions by
substituting back into the original ODEs.
(a)
du
dx
= 5u + 3v
dv
dx
= u + 7v
(b)
du
dx
= 6u + 3v
dv
dx
= 4u v
(c)
du
dx
= 4u 2v
dv
dx
= u + 3v
(d)
du
dx
= 8u + 4v
dv
dx
= 7u 3v
5. Solve each of the coupled ODEs of the previous question by way of eigenvectors and
eigenvalues.
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory class 10
Limits
1. At which points are the following functions discontinuous (if any)? Assume the domain
for each function to be R or R
2
.
(a) f(x) = sin(x) (b) g(x) = (2 x)/(2 + x)
(c) h(x) = log x (d) p(x) = (1 + 2x x
2
)/(1 + 2x +x
2
)
(e) r(x, y) = tan(x + y) (f) s(x, y) = (x y)
2
/(x + y)
2
(g) t(u, v) = (1 + u + u
2
)/(1 + v + v
2
) (h) w(u, v) = exp(u
2
v
2
)
2. Use your calculator to estimate the following limits.
(a) lim
x0
sin(x)
x
(b) lim
x1
1 + x
1 x
(c) lim
(x,y)(0,0)
sin(x + y)
x + y
(d) lim
(x,y)(1,1)
(x + y 1)
2
(x y + 1)
2
(e) lim
(x,y)(1,0)
x
2
y
2
1
x
2
+ y
2
1
(f) lim
(x,y)(0,0)
1 exp(x
2
y
2
)
xy
Partial Derivatives
3. Evaluate the rst partial derivatives for each of the following functions
(a) f(x, y) = cos(x) cos(y) (b) f(x, y) = sin(xy)
(c) f(x, y) = log(1 + x)/ log(1 + y) (d) f(x, y) = (x + y)/(x y)
(e) f(x, y) = xy (f) f(u, v) = uv(1 u
2
v
2
)
4. For the function f(x, y) = y
2
sin(x) verify that

x
_
f
y
_
=

y
_
f
x
_
School of Mathematical Sciences Monash University
Chain Rule
5. Given f(x, y) = 2x
2
+ 4y 2 and x(s) = 3s, y(s) = 2s
2
compute df/ds by direct
substitution (i.e. rst construct f(s)) and also by the chain rule.
6. Given f(x, y) = 2xy and x(r, ) = r cos , y(r, ) = r sin compute f/x, f/y,
f/r and f/,
7. Let f = f(x, y) be an arbitrary function of (x, y). Using the same transformation as in
the previous question express

2
f
x
2
+

2
f
y
2
in terms of partial derivatives of f in r and . This is a long and tedious question have
fun!
Directional derivatives
8. Compute df/ds for the function f(x, y) = xy +x +y along the curve x(s) = r cos(s/r),
y(s) = r sin(s/r). Also, verify that (dx/s) i

+ (dy/ds)j

is a unit vector.
9. Compute the directional derivative for each for the following functions in the stated
direction. Be sure that you use a unit vector!
(a) f(x, y) = 2x + 3y at (1, 2), t

= (3 i

+ 4j

)/5
(b) g(x, y) = sin(x) cos(y) at (/4, /4), t

= ( i

+ j

)/

2
(c) h(x, y, z) = log(x
2
+ y
2
+ z
2
) at (1, 0, 1), t

= i

+ j

(d) q(x, y, z) = 4x
2
3y
3
+ 2z
2
at (0, 1, 2), t

= 2 i

3j

+ k

(e) r(x, y, z) = z exp(2xy) at (1, 1, 1), t

= i

3j

+ 2k

(f) w(x, y, z) =
_
1 x
2
y
2
z
2
at (0.5, 0.5, 0.5), t

= 2 i

+ k

Tangent planes
10. Compute the tangent plane

f approximation for each of the following functions at the
stated point.
(a) f(x, y) = 2x + 3y at (1, 2)
(b) g(x, y) = sin(x) cos(y) at (/4, /4)
(c) h(x, y, z) = log(x
2
+ y
2
+ z
2
) at (1, 0, 1)
(d) q(x, y, z) = 4x
2
3y
3
+ 2z
2
at (0, 1, 2)
(e) r(x, y, z) = z exp(2xy) at (1, 1, 1)
(f) w(x, y, z) =
_
1 x
2
y
2
z
2
at (0.5, 0.5, 0.5)
26-Jul-2013 217
School of Mathematical Sciences Monash University
11. Use the result from the previous question to estimate the function at the stated points.
Compare your estimate with that given by a calculator.
(a) f(x, y) at (1.1, 1.9) (b) g(x, y) at (3/16, 5/16)
(c) h(x, y, z) at (0.8, 0.1, 0.9) (d) q(x, y, z) at (0.1, 1.1, 1.9)
(e) r(x, y, z) at (0.8, 1.2, 1.1) (f) w(x, y, z) at (0.6, 0.4, 0.6)
12. This is more a question on theory rather than being a pure number question. It is thus
not examinable.
Consider a function f = f(x, y) and its tangent plane approximation

f at some point
P. Both of these may be drawn as surfaces in 3-dimensional space. You might ask
How can I compute the normal vector to the surface for f at the point P? And that is
exactly what we will do in this question.
Construct

f at P (i.e write down the standard formula for

f). Draw this as a surface in
the 3-dimensional space. This surface is a at plane tangent to the surface for f at P
(hence the name, tangent plane).
Given your equation for the plane, write down a 3-vector normal to this plane. Hence
deduce the normal to the surface for the function f = f(x, y) at P.
13. Generalise your result from the previous question to surfaces of the form 0 = g(x, y, z).
This question is also a non-examinable extension. But it is fun! (agreed?).
Maxima and Minima
14. Find all of the extrema (if any) for each of the following functions (you do not need to
charactise the extrema).
(a) f(x, y) = 4 x
2
y
2
(b) g(x, y) = xy exp(x
2
y
2
)
(c) h(x, y) = x x
3
+ y
2
(d) p(x, y) = (2 x
2
) exp(y)
(e) q(x, y, z) = 4x
2
+ 3y
2
+ z
2
(f) r(x, y, z) = arctan((x1)
2
+y
2
+z
2
)
26-Jul-2013 218
SCHOOL OF MATHEMATICAL SCIENCES
ENG1091
Mathematics for Engineering
Laboratory class 10 Solutions
Limits
1. At which points are the following functions discontinuous (if any)? Assume the domain
for each function to be R or R
2
.
(a) None (b) x = 2
(c) x = 0 (d) x = 1
(e) x + y = /2, 3/2, 5/2 (f) x + y = 0
(g) None (h) None
2. Use your calculator to estimate the following limits.
(a) 1 (b)
(c) 1 (d) 1
(e) No unique limit, try limits along
the axes.
(f) 0
Partial Derivatives
3. Evaluate the rst partial derivatives for each of the following functions
(a)
f
x
= sin(x) cos(y)
f
y
= cos(x) sin(y)
(b)
f
x
= y cos(xy)
f
y
= x cos(xy)
(c)
f
x
=
1
(1 + x) log(1 + y)
f
y
=
log(1 + x)
(1 + y) log
2
(1 + y)
(d)
f
x
=
2y
(x y)
2
f
y
=
2x
(x y)
2
(e)
f
x
= y
f
y
= x
(f)
f
u
= v(1 3u
2
v
2
)
f
v
= u(1 u
2
3v
2
)
School of Mathematical Sciences Monash University
Chain Rule
5. df/ds = 52s
6. f/x = 2y, f/y = 2x, f/r = 4r cos sin , f/ = 2r
2
(cos
2
sin
2
),
7. This is not an easy question, two chocolate frogs if you got it right!

2
f
x
2
+

2
f
y
2
=

2
f
r
2
+
1
r
f

+
1
r
2

2
f

2
Directional derivatives
8. df/ds = r
_
cos
2
(s/r) sin
2
(s/r)
_
sin(s/r) + cos(s/r).
9. (a) 18/5 (b) 0
(c) 0 (d) 35/

14
(e) 2 exp(2)/

14 (f) 2/

6
Tangent planes
10. (a)

f(x, y) = 8 + 2(x 1) + 3(y 2)
(b)

f(x, y) = (1/2) + (1/2)(x /4) (1/2)(y /4)
(c)

f(x, y, z) = log 2 + (x 1) + (z 1)
(d)

f(x, y, z) = 5 9(y 1) + 8(z 2)
(e)

f(x, y, z) = exp(2)(1 + 2(x 1) + 2(y 1) + (z + 1))
(f)

f(x, y, z) = (1/2) (x (1/2)) (y (1/2)) (z (1/2))
11. The calculators answer is in brackets.
(a) 7.9 (7.900) (b) 0.304 (0.2397)
(c) 0.393 (0.3784) (d) 3.7 (3.267)
(e) -0.149 (-0.1613) (f) 0.4 (0.3464)
12. This question is not examinable.
For a surface written in the form z = f(x, y) the vector
N =
_
f
x
_
i

+
_
f
y
_
j

is normal to the surface.


26-Jul-2013 220
School of Mathematical Sciences Monash University
13. This question is not examinable.
For a surface written in the form 0 = g(x, y, z) the vector
N = g =
_
g
x
_
i

+
_
g
y
_
j

+
_
g
z
_
k

is normal to the surface.


Maxima and Minima
14. (a) (0, 0) (b) (0, 0) and the four points (1/

2, 1/

2)
(c) (1/

3, 0) (d) None
(e) (0, 0, 0) (f) (1, 0, 0)
26-Jul-2013 221

You might also like